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/uprobes.h>
27#include <linux/string.h>
28#include <linux/delay.h>
29#include <linux/errno.h>
30#include <linux/kexec.h>
31#include <linux/sched.h>
32#include <linux/timer.h>
33#include <linux/init.h>
34#include <linux/bug.h>
35#include <linux/nmi.h>
36#include <linux/mm.h>
37#include <linux/smp.h>
38#include <linux/io.h>
39
40#ifdef CONFIG_EISA
41#include <linux/ioport.h>
42#include <linux/eisa.h>
43#endif
44
45#if defined(CONFIG_EDAC)
46#include <linux/edac.h>
47#endif
48
49#include <asm/kmemcheck.h>
50#include <asm/stacktrace.h>
51#include <asm/processor.h>
52#include <asm/debugreg.h>
53#include <linux/atomic.h>
54#include <asm/ftrace.h>
55#include <asm/traps.h>
56#include <asm/desc.h>
57#include <asm/i387.h>
58#include <asm/fpu-internal.h>
59#include <asm/mce.h>
60#include <asm/fixmap.h>
61#include <asm/mach_traps.h>
62#include <asm/alternative.h>
63
64#ifdef CONFIG_X86_64
65#include <asm/x86_init.h>
66#include <asm/pgalloc.h>
67#include <asm/proto.h>
68
69/* No need to be aligned, but done to keep all IDTs defined the same way. */
70gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
71#else
72#include <asm/processor-flags.h>
73#include <asm/setup.h>
74
75asmlinkage int system_call(void);
76#endif
77
78/* Must be page-aligned because the real IDT is used in a fixmap. */
79gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
80
81DECLARE_BITMAP(used_vectors, NR_VECTORS);
82EXPORT_SYMBOL_GPL(used_vectors);
83
84static inline void conditional_sti(struct pt_regs *regs)
85{
86	if (regs->flags & X86_EFLAGS_IF)
87		local_irq_enable();
88}
89
90static inline void preempt_conditional_sti(struct pt_regs *regs)
91{
92	preempt_count_inc();
93	if (regs->flags & X86_EFLAGS_IF)
94		local_irq_enable();
95}
96
97static inline void conditional_cli(struct pt_regs *regs)
98{
99	if (regs->flags & X86_EFLAGS_IF)
100		local_irq_disable();
101}
102
103static inline void preempt_conditional_cli(struct pt_regs *regs)
104{
105	if (regs->flags & X86_EFLAGS_IF)
106		local_irq_disable();
107	preempt_count_dec();
108}
109
110static nokprobe_inline int
111do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
112		  struct pt_regs *regs,	long error_code)
113{
114#ifdef CONFIG_X86_32
115	if (regs->flags & X86_VM_MASK) {
116		/*
117		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
118		 * On nmi (interrupt 2), do_trap should not be called.
119		 */
120		if (trapnr < X86_TRAP_UD) {
121			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
122						error_code, trapnr))
123				return 0;
124		}
125		return -1;
126	}
127#endif
128	if (!user_mode(regs)) {
129		if (!fixup_exception(regs)) {
130			tsk->thread.error_code = error_code;
131			tsk->thread.trap_nr = trapnr;
132			die(str, regs, error_code);
133		}
134		return 0;
135	}
136
137	return -1;
138}
139
140static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
141				siginfo_t *info)
142{
143	unsigned long siaddr;
144	int sicode;
145
146	switch (trapnr) {
147	default:
148		return SEND_SIG_PRIV;
149
150	case X86_TRAP_DE:
151		sicode = FPE_INTDIV;
152		siaddr = uprobe_get_trap_addr(regs);
153		break;
154	case X86_TRAP_UD:
155		sicode = ILL_ILLOPN;
156		siaddr = uprobe_get_trap_addr(regs);
157		break;
158	case X86_TRAP_AC:
159		sicode = BUS_ADRALN;
160		siaddr = 0;
161		break;
162	}
163
164	info->si_signo = signr;
165	info->si_errno = 0;
166	info->si_code = sicode;
167	info->si_addr = (void __user *)siaddr;
168	return info;
169}
170
171static void
172do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
173	long error_code, siginfo_t *info)
174{
175	struct task_struct *tsk = current;
176
177
178	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
179		return;
180	/*
181	 * We want error_code and trap_nr set for userspace faults and
182	 * kernelspace faults which result in die(), but not
183	 * kernelspace faults which are fixed up.  die() gives the
184	 * process no chance to handle the signal and notice the
185	 * kernel fault information, so that won't result in polluting
186	 * the information about previously queued, but not yet
187	 * delivered, faults.  See also do_general_protection below.
188	 */
189	tsk->thread.error_code = error_code;
190	tsk->thread.trap_nr = trapnr;
191
192#ifdef CONFIG_X86_64
193	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
194	    printk_ratelimit()) {
195		pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
196			tsk->comm, tsk->pid, str,
197			regs->ip, regs->sp, error_code);
198		print_vma_addr(" in ", regs->ip);
199		pr_cont("\n");
200	}
201#endif
202
203	force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
204}
205NOKPROBE_SYMBOL(do_trap);
206
207static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
208			  unsigned long trapnr, int signr)
209{
210	enum ctx_state prev_state = exception_enter();
211	siginfo_t info;
212
213	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
214			NOTIFY_STOP) {
215		conditional_sti(regs);
216		do_trap(trapnr, signr, str, regs, error_code,
217			fill_trap_info(regs, signr, trapnr, &info));
218	}
219
220	exception_exit(prev_state);
221}
222
223#define DO_ERROR(trapnr, signr, str, name)				\
224dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
225{									\
226	do_error_trap(regs, error_code, str, trapnr, signr);		\
227}
228
229DO_ERROR(X86_TRAP_DE,     SIGFPE,  "divide error",		divide_error)
230DO_ERROR(X86_TRAP_OF,     SIGSEGV, "overflow",			overflow)
231DO_ERROR(X86_TRAP_BR,     SIGSEGV, "bounds",			bounds)
232DO_ERROR(X86_TRAP_UD,     SIGILL,  "invalid opcode",		invalid_op)
233DO_ERROR(X86_TRAP_OLD_MF, SIGFPE,  "coprocessor segment overrun",coprocessor_segment_overrun)
234DO_ERROR(X86_TRAP_TS,     SIGSEGV, "invalid TSS",		invalid_TSS)
235DO_ERROR(X86_TRAP_NP,     SIGBUS,  "segment not present",	segment_not_present)
236DO_ERROR(X86_TRAP_SS,     SIGBUS,  "stack segment",		stack_segment)
237DO_ERROR(X86_TRAP_AC,     SIGBUS,  "alignment check",		alignment_check)
238
239#ifdef CONFIG_X86_64
240/* Runs on IST stack */
241dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
242{
243	static const char str[] = "double fault";
244	struct task_struct *tsk = current;
245
246#ifdef CONFIG_X86_ESPFIX64
247	extern unsigned char native_irq_return_iret[];
248
249	/*
250	 * If IRET takes a non-IST fault on the espfix64 stack, then we
251	 * end up promoting it to a doublefault.  In that case, modify
252	 * the stack to make it look like we just entered the #GP
253	 * handler from user space, similar to bad_iret.
254	 */
255	if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
256		regs->cs == __KERNEL_CS &&
257		regs->ip == (unsigned long)native_irq_return_iret)
258	{
259		struct pt_regs *normal_regs = task_pt_regs(current);
260
261		/* Fake a #GP(0) from userspace. */
262		memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
263		normal_regs->orig_ax = 0;  /* Missing (lost) #GP error code */
264		regs->ip = (unsigned long)general_protection;
265		regs->sp = (unsigned long)&normal_regs->orig_ax;
266		return;
267	}
268#endif
269
270	exception_enter();
271	/* Return not checked because double check cannot be ignored */
272	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
273
274	tsk->thread.error_code = error_code;
275	tsk->thread.trap_nr = X86_TRAP_DF;
276
277#ifdef CONFIG_DOUBLEFAULT
278	df_debug(regs, error_code);
279#endif
280	/*
281	 * This is always a kernel trap and never fixable (and thus must
282	 * never return).
283	 */
284	for (;;)
285		die(str, regs, error_code);
286}
287#endif
288
289dotraplinkage void
290do_general_protection(struct pt_regs *regs, long error_code)
291{
292	struct task_struct *tsk;
293	enum ctx_state prev_state;
294
295	prev_state = exception_enter();
296	conditional_sti(regs);
297
298#ifdef CONFIG_X86_32
299	if (regs->flags & X86_VM_MASK) {
300		local_irq_enable();
301		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
302		goto exit;
303	}
304#endif
305
306	tsk = current;
307	if (!user_mode(regs)) {
308		if (fixup_exception(regs))
309			goto exit;
310
311		tsk->thread.error_code = error_code;
312		tsk->thread.trap_nr = X86_TRAP_GP;
313		if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
314			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
315			die("general protection fault", regs, error_code);
316		goto exit;
317	}
318
319	tsk->thread.error_code = error_code;
320	tsk->thread.trap_nr = X86_TRAP_GP;
321
322	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
323			printk_ratelimit()) {
324		pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
325			tsk->comm, task_pid_nr(tsk),
326			regs->ip, regs->sp, error_code);
327		print_vma_addr(" in ", regs->ip);
328		pr_cont("\n");
329	}
330
331	force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
332exit:
333	exception_exit(prev_state);
334}
335NOKPROBE_SYMBOL(do_general_protection);
336
337/* May run on IST stack. */
338dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
339{
340	enum ctx_state prev_state;
341
342#ifdef CONFIG_DYNAMIC_FTRACE
343	/*
344	 * ftrace must be first, everything else may cause a recursive crash.
345	 * See note by declaration of modifying_ftrace_code in ftrace.c
346	 */
347	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
348	    ftrace_int3_handler(regs))
349		return;
350#endif
351	if (poke_int3_handler(regs))
352		return;
353
354	prev_state = exception_enter();
355#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
356	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
357				SIGTRAP) == NOTIFY_STOP)
358		goto exit;
359#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
360
361#ifdef CONFIG_KPROBES
362	if (kprobe_int3_handler(regs))
363		goto exit;
364#endif
365
366	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
367			SIGTRAP) == NOTIFY_STOP)
368		goto exit;
369
370	/*
371	 * Let others (NMI) know that the debug stack is in use
372	 * as we may switch to the interrupt stack.
373	 */
374	debug_stack_usage_inc();
375	preempt_conditional_sti(regs);
376	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
377	preempt_conditional_cli(regs);
378	debug_stack_usage_dec();
379exit:
380	exception_exit(prev_state);
381}
382NOKPROBE_SYMBOL(do_int3);
383
384#ifdef CONFIG_X86_64
385/*
386 * Help handler running on IST stack to switch back to user stack
387 * for scheduling or signal handling. The actual stack switch is done in
388 * entry.S
389 */
390asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
391{
392	struct pt_regs *regs = eregs;
393	/* Did already sync */
394	if (eregs == (struct pt_regs *)eregs->sp)
395		;
396	/* Exception from user space */
397	else if (user_mode(eregs))
398		regs = task_pt_regs(current);
399	/*
400	 * Exception from kernel and interrupts are enabled. Move to
401	 * kernel process stack.
402	 */
403	else if (eregs->flags & X86_EFLAGS_IF)
404		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
405	if (eregs != regs)
406		*regs = *eregs;
407	return regs;
408}
409NOKPROBE_SYMBOL(sync_regs);
410
411struct bad_iret_stack {
412	void *error_entry_ret;
413	struct pt_regs regs;
414};
415
416asmlinkage __visible notrace
417struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
418{
419	/*
420	 * This is called from entry_64.S early in handling a fault
421	 * caused by a bad iret to user mode.  To handle the fault
422	 * correctly, we want move our stack frame to task_pt_regs
423	 * and we want to pretend that the exception came from the
424	 * iret target.
425	 */
426	struct bad_iret_stack *new_stack =
427		container_of(task_pt_regs(current),
428			     struct bad_iret_stack, regs);
429
430	/* Copy the IRET target to the new stack. */
431	memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
432
433	/* Copy the remainder of the stack from the current stack. */
434	memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
435
436	BUG_ON(!user_mode_vm(&new_stack->regs));
437	return new_stack;
438}
439NOKPROBE_SYMBOL(fixup_bad_iret);
440#endif
441
442/*
443 * Our handling of the processor debug registers is non-trivial.
444 * We do not clear them on entry and exit from the kernel. Therefore
445 * it is possible to get a watchpoint trap here from inside the kernel.
446 * However, the code in ./ptrace.c has ensured that the user can
447 * only set watchpoints on userspace addresses. Therefore the in-kernel
448 * watchpoint trap can only occur in code which is reading/writing
449 * from user space. Such code must not hold kernel locks (since it
450 * can equally take a page fault), therefore it is safe to call
451 * force_sig_info even though that claims and releases locks.
452 *
453 * Code in ./signal.c ensures that the debug control register
454 * is restored before we deliver any signal, and therefore that
455 * user code runs with the correct debug control register even though
456 * we clear it here.
457 *
458 * Being careful here means that we don't have to be as careful in a
459 * lot of more complicated places (task switching can be a bit lazy
460 * about restoring all the debug state, and ptrace doesn't have to
461 * find every occurrence of the TF bit that could be saved away even
462 * by user code)
463 *
464 * May run on IST stack.
465 */
466dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
467{
468	struct task_struct *tsk = current;
469	enum ctx_state prev_state;
470	int user_icebp = 0;
471	unsigned long dr6;
472	int si_code;
473
474	prev_state = exception_enter();
475
476	get_debugreg(dr6, 6);
477
478	/* Filter out all the reserved bits which are preset to 1 */
479	dr6 &= ~DR6_RESERVED;
480
481	/*
482	 * If dr6 has no reason to give us about the origin of this trap,
483	 * then it's very likely the result of an icebp/int01 trap.
484	 * User wants a sigtrap for that.
485	 */
486	if (!dr6 && user_mode(regs))
487		user_icebp = 1;
488
489	/* Catch kmemcheck conditions first of all! */
490	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
491		goto exit;
492
493	/* DR6 may or may not be cleared by the CPU */
494	set_debugreg(0, 6);
495
496	/*
497	 * The processor cleared BTF, so don't mark that we need it set.
498	 */
499	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
500
501	/* Store the virtualized DR6 value */
502	tsk->thread.debugreg6 = dr6;
503
504#ifdef CONFIG_KPROBES
505	if (kprobe_debug_handler(regs))
506		goto exit;
507#endif
508
509	if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
510							SIGTRAP) == NOTIFY_STOP)
511		goto exit;
512
513	/*
514	 * Let others (NMI) know that the debug stack is in use
515	 * as we may switch to the interrupt stack.
516	 */
517	debug_stack_usage_inc();
518
519	/* It's safe to allow irq's after DR6 has been saved */
520	preempt_conditional_sti(regs);
521
522	if (regs->flags & X86_VM_MASK) {
523		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
524					X86_TRAP_DB);
525		preempt_conditional_cli(regs);
526		debug_stack_usage_dec();
527		goto exit;
528	}
529
530	/*
531	 * Single-stepping through system calls: ignore any exceptions in
532	 * kernel space, but re-enable TF when returning to user mode.
533	 *
534	 * We already checked v86 mode above, so we can check for kernel mode
535	 * by just checking the CPL of CS.
536	 */
537	if ((dr6 & DR_STEP) && !user_mode(regs)) {
538		tsk->thread.debugreg6 &= ~DR_STEP;
539		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
540		regs->flags &= ~X86_EFLAGS_TF;
541	}
542	si_code = get_si_code(tsk->thread.debugreg6);
543	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
544		send_sigtrap(tsk, regs, error_code, si_code);
545	preempt_conditional_cli(regs);
546	debug_stack_usage_dec();
547
548exit:
549	exception_exit(prev_state);
550}
551NOKPROBE_SYMBOL(do_debug);
552
553/*
554 * Note that we play around with the 'TS' bit in an attempt to get
555 * the correct behaviour even in the presence of the asynchronous
556 * IRQ13 behaviour
557 */
558static void math_error(struct pt_regs *regs, int error_code, int trapnr)
559{
560	struct task_struct *task = current;
561	siginfo_t info;
562	unsigned short err;
563	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
564						"simd exception";
565
566	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
567		return;
568	conditional_sti(regs);
569
570	if (!user_mode_vm(regs))
571	{
572		if (!fixup_exception(regs)) {
573			task->thread.error_code = error_code;
574			task->thread.trap_nr = trapnr;
575			die(str, regs, error_code);
576		}
577		return;
578	}
579
580	/*
581	 * Save the info for the exception handler and clear the error.
582	 */
583	save_init_fpu(task);
584	task->thread.trap_nr = trapnr;
585	task->thread.error_code = error_code;
586	info.si_signo = SIGFPE;
587	info.si_errno = 0;
588	info.si_addr = (void __user *)uprobe_get_trap_addr(regs);
589	if (trapnr == X86_TRAP_MF) {
590		unsigned short cwd, swd;
591		/*
592		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
593		 * status.  0x3f is the exception bits in these regs, 0x200 is the
594		 * C1 reg you need in case of a stack fault, 0x040 is the stack
595		 * fault bit.  We should only be taking one exception at a time,
596		 * so if this combination doesn't produce any single exception,
597		 * then we have a bad program that isn't synchronizing its FPU usage
598		 * and it will suffer the consequences since we won't be able to
599		 * fully reproduce the context of the exception
600		 */
601		cwd = get_fpu_cwd(task);
602		swd = get_fpu_swd(task);
603
604		err = swd & ~cwd;
605	} else {
606		/*
607		 * The SIMD FPU exceptions are handled a little differently, as there
608		 * is only a single status/control register.  Thus, to determine which
609		 * unmasked exception was caught we must mask the exception mask bits
610		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
611		 */
612		unsigned short mxcsr = get_fpu_mxcsr(task);
613		err = ~(mxcsr >> 7) & mxcsr;
614	}
615
616	if (err & 0x001) {	/* Invalid op */
617		/*
618		 * swd & 0x240 == 0x040: Stack Underflow
619		 * swd & 0x240 == 0x240: Stack Overflow
620		 * User must clear the SF bit (0x40) if set
621		 */
622		info.si_code = FPE_FLTINV;
623	} else if (err & 0x004) { /* Divide by Zero */
624		info.si_code = FPE_FLTDIV;
625	} else if (err & 0x008) { /* Overflow */
626		info.si_code = FPE_FLTOVF;
627	} else if (err & 0x012) { /* Denormal, Underflow */
628		info.si_code = FPE_FLTUND;
629	} else if (err & 0x020) { /* Precision */
630		info.si_code = FPE_FLTRES;
631	} else {
632		/*
633		 * If we're using IRQ 13, or supposedly even some trap
634		 * X86_TRAP_MF implementations, it's possible
635		 * we get a spurious trap, which is not an error.
636		 */
637		return;
638	}
639	force_sig_info(SIGFPE, &info, task);
640}
641
642dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
643{
644	enum ctx_state prev_state;
645
646	prev_state = exception_enter();
647	math_error(regs, error_code, X86_TRAP_MF);
648	exception_exit(prev_state);
649}
650
651dotraplinkage void
652do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
653{
654	enum ctx_state prev_state;
655
656	prev_state = exception_enter();
657	math_error(regs, error_code, X86_TRAP_XF);
658	exception_exit(prev_state);
659}
660
661dotraplinkage void
662do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
663{
664	conditional_sti(regs);
665#if 0
666	/* No need to warn about this any longer. */
667	pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
668#endif
669}
670
671asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
672{
673}
674
675asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
676{
677}
678
679/*
680 * 'math_state_restore()' saves the current math information in the
681 * old math state array, and gets the new ones from the current task
682 *
683 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
684 * Don't touch unless you *really* know how it works.
685 *
686 * Must be called with kernel preemption disabled (eg with local
687 * local interrupts as in the case of do_device_not_available).
688 */
689void math_state_restore(void)
690{
691	struct task_struct *tsk = current;
692
693	if (!tsk_used_math(tsk)) {
694		local_irq_enable();
695		/*
696		 * does a slab alloc which can sleep
697		 */
698		if (init_fpu(tsk)) {
699			/*
700			 * ran out of memory!
701			 */
702			do_group_exit(SIGKILL);
703			return;
704		}
705		local_irq_disable();
706	}
707
708	__thread_fpu_begin(tsk);
709
710	/*
711	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
712	 */
713	if (unlikely(restore_fpu_checking(tsk))) {
714		drop_init_fpu(tsk);
715		force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
716		return;
717	}
718
719	tsk->thread.fpu_counter++;
720}
721EXPORT_SYMBOL_GPL(math_state_restore);
722
723dotraplinkage void
724do_device_not_available(struct pt_regs *regs, long error_code)
725{
726	enum ctx_state prev_state;
727
728	prev_state = exception_enter();
729	BUG_ON(use_eager_fpu());
730
731#ifdef CONFIG_MATH_EMULATION
732	if (read_cr0() & X86_CR0_EM) {
733		struct math_emu_info info = { };
734
735		conditional_sti(regs);
736
737		info.regs = regs;
738		math_emulate(&info);
739		exception_exit(prev_state);
740		return;
741	}
742#endif
743	math_state_restore(); /* interrupts still off */
744#ifdef CONFIG_X86_32
745	conditional_sti(regs);
746#endif
747	exception_exit(prev_state);
748}
749NOKPROBE_SYMBOL(do_device_not_available);
750
751#ifdef CONFIG_X86_32
752dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
753{
754	siginfo_t info;
755	enum ctx_state prev_state;
756
757	prev_state = exception_enter();
758	local_irq_enable();
759
760	info.si_signo = SIGILL;
761	info.si_errno = 0;
762	info.si_code = ILL_BADSTK;
763	info.si_addr = NULL;
764	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
765			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
766		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
767			&info);
768	}
769	exception_exit(prev_state);
770}
771#endif
772
773/* Set of traps needed for early debugging. */
774void __init early_trap_init(void)
775{
776	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
777	/* int3 can be called from all */
778	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
779#ifdef CONFIG_X86_32
780	set_intr_gate(X86_TRAP_PF, page_fault);
781#endif
782	load_idt(&idt_descr);
783}
784
785void __init early_trap_pf_init(void)
786{
787#ifdef CONFIG_X86_64
788	set_intr_gate(X86_TRAP_PF, page_fault);
789#endif
790}
791
792void __init trap_init(void)
793{
794	int i;
795
796#ifdef CONFIG_EISA
797	void __iomem *p = early_ioremap(0x0FFFD9, 4);
798
799	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
800		EISA_bus = 1;
801	early_iounmap(p, 4);
802#endif
803
804	set_intr_gate(X86_TRAP_DE, divide_error);
805	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
806	/* int4 can be called from all */
807	set_system_intr_gate(X86_TRAP_OF, &overflow);
808	set_intr_gate(X86_TRAP_BR, bounds);
809	set_intr_gate(X86_TRAP_UD, invalid_op);
810	set_intr_gate(X86_TRAP_NM, device_not_available);
811#ifdef CONFIG_X86_32
812	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
813#else
814	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
815#endif
816	set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
817	set_intr_gate(X86_TRAP_TS, invalid_TSS);
818	set_intr_gate(X86_TRAP_NP, segment_not_present);
819	set_intr_gate(X86_TRAP_SS, stack_segment);
820	set_intr_gate(X86_TRAP_GP, general_protection);
821	set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
822	set_intr_gate(X86_TRAP_MF, coprocessor_error);
823	set_intr_gate(X86_TRAP_AC, alignment_check);
824#ifdef CONFIG_X86_MCE
825	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
826#endif
827	set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
828
829	/* Reserve all the builtin and the syscall vector: */
830	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
831		set_bit(i, used_vectors);
832
833#ifdef CONFIG_IA32_EMULATION
834	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
835	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
836#endif
837
838#ifdef CONFIG_X86_32
839	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
840	set_bit(SYSCALL_VECTOR, used_vectors);
841#endif
842
843	/*
844	 * Set the IDT descriptor to a fixed read-only location, so that the
845	 * "sidt" instruction will not leak the location of the kernel, and
846	 * to defend the IDT against arbitrary memory write vulnerabilities.
847	 * It will be reloaded in cpu_init() */
848	__set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
849	idt_descr.address = fix_to_virt(FIX_RO_IDT);
850
851	/*
852	 * Should be a barrier for any external CPU state:
853	 */
854	cpu_init();
855
856	x86_init.irqs.trap_init();
857
858#ifdef CONFIG_X86_64
859	memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
860	set_nmi_gate(X86_TRAP_DB, &debug);
861	set_nmi_gate(X86_TRAP_BP, &int3);
862#endif
863}
864