xref: /openbmc/linux/arch/x86/kernel/traps.c (revision 5dc3055879b8f659f62abb7c3d1eaa4d02e36d65)
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 <asm/atomic.h>
53 #include <asm/system.h>
54 #include <asm/traps.h>
55 #include <asm/desc.h>
56 #include <asm/i387.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 
84 static int ignore_nmis;
85 
86 int unknown_nmi_panic;
87 
88 static inline void conditional_sti(struct pt_regs *regs)
89 {
90 	if (regs->flags & X86_EFLAGS_IF)
91 		local_irq_enable();
92 }
93 
94 static inline void preempt_conditional_sti(struct pt_regs *regs)
95 {
96 	inc_preempt_count();
97 	if (regs->flags & X86_EFLAGS_IF)
98 		local_irq_enable();
99 }
100 
101 static inline void conditional_cli(struct pt_regs *regs)
102 {
103 	if (regs->flags & X86_EFLAGS_IF)
104 		local_irq_disable();
105 }
106 
107 static inline void preempt_conditional_cli(struct pt_regs *regs)
108 {
109 	if (regs->flags & X86_EFLAGS_IF)
110 		local_irq_disable();
111 	dec_preempt_count();
112 }
113 
114 static void __kprobes
115 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
116 	long error_code, siginfo_t *info)
117 {
118 	struct task_struct *tsk = current;
119 
120 #ifdef CONFIG_X86_32
121 	if (regs->flags & X86_VM_MASK) {
122 		/*
123 		 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
124 		 * On nmi (interrupt 2), do_trap should not be called.
125 		 */
126 		if (trapnr < 6)
127 			goto vm86_trap;
128 		goto trap_signal;
129 	}
130 #endif
131 
132 	if (!user_mode(regs))
133 		goto kernel_trap;
134 
135 #ifdef CONFIG_X86_32
136 trap_signal:
137 #endif
138 	/*
139 	 * We want error_code and trap_no set for userspace faults and
140 	 * kernelspace faults which result in die(), but not
141 	 * kernelspace faults which are fixed up.  die() gives the
142 	 * process no chance to handle the signal and notice the
143 	 * kernel fault information, so that won't result in polluting
144 	 * the information about previously queued, but not yet
145 	 * delivered, faults.  See also do_general_protection below.
146 	 */
147 	tsk->thread.error_code = error_code;
148 	tsk->thread.trap_no = trapnr;
149 
150 #ifdef CONFIG_X86_64
151 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
152 	    printk_ratelimit()) {
153 		printk(KERN_INFO
154 		       "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
155 		       tsk->comm, tsk->pid, str,
156 		       regs->ip, regs->sp, error_code);
157 		print_vma_addr(" in ", regs->ip);
158 		printk("\n");
159 	}
160 #endif
161 
162 	if (info)
163 		force_sig_info(signr, info, tsk);
164 	else
165 		force_sig(signr, tsk);
166 	return;
167 
168 kernel_trap:
169 	if (!fixup_exception(regs)) {
170 		tsk->thread.error_code = error_code;
171 		tsk->thread.trap_no = trapnr;
172 		die(str, regs, error_code);
173 	}
174 	return;
175 
176 #ifdef CONFIG_X86_32
177 vm86_trap:
178 	if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
179 						error_code, trapnr))
180 		goto trap_signal;
181 	return;
182 #endif
183 }
184 
185 #define DO_ERROR(trapnr, signr, str, name)				\
186 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
187 {									\
188 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
189 							== NOTIFY_STOP)	\
190 		return;							\
191 	conditional_sti(regs);						\
192 	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
193 }
194 
195 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)		\
196 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
197 {									\
198 	siginfo_t info;							\
199 	info.si_signo = signr;						\
200 	info.si_errno = 0;						\
201 	info.si_code = sicode;						\
202 	info.si_addr = (void __user *)siaddr;				\
203 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
204 							== NOTIFY_STOP)	\
205 		return;							\
206 	conditional_sti(regs);						\
207 	do_trap(trapnr, signr, str, regs, error_code, &info);		\
208 }
209 
210 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
211 DO_ERROR(4, SIGSEGV, "overflow", overflow)
212 DO_ERROR(5, SIGSEGV, "bounds", bounds)
213 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
214 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
215 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
216 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
217 #ifdef CONFIG_X86_32
218 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
219 #endif
220 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
221 
222 #ifdef CONFIG_X86_64
223 /* Runs on IST stack */
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 			12, SIGBUS) == NOTIFY_STOP)
228 		return;
229 	preempt_conditional_sti(regs);
230 	do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
231 	preempt_conditional_cli(regs);
232 }
233 
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, 8, SIGSEGV);
241 
242 	tsk->thread.error_code = error_code;
243 	tsk->thread.trap_no = 8;
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
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_no = 13;
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_no = 13;
299 	if (notify_die(DIE_GPF, "general protection fault", regs,
300 				error_code, 13, SIGSEGV) == NOTIFY_STOP)
301 		return;
302 	die("general protection fault", regs, error_code);
303 }
304 
305 static int __init setup_unknown_nmi_panic(char *str)
306 {
307 	unknown_nmi_panic = 1;
308 	return 1;
309 }
310 __setup("unknown_nmi_panic", setup_unknown_nmi_panic);
311 
312 static notrace __kprobes void
313 mem_parity_error(unsigned char reason, struct pt_regs *regs)
314 {
315 	printk(KERN_EMERG
316 		"Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
317 			reason, smp_processor_id());
318 
319 	printk(KERN_EMERG
320 		"You have some hardware problem, likely on the PCI bus.\n");
321 
322 #if defined(CONFIG_EDAC)
323 	if (edac_handler_set()) {
324 		edac_atomic_assert_error();
325 		return;
326 	}
327 #endif
328 
329 	if (panic_on_unrecovered_nmi)
330 		panic("NMI: Not continuing");
331 
332 	printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
333 
334 	/* Clear and disable the memory parity error line. */
335 	reason = (reason & 0xf) | 4;
336 	outb(reason, 0x61);
337 }
338 
339 static notrace __kprobes void
340 io_check_error(unsigned char reason, struct pt_regs *regs)
341 {
342 	unsigned long i;
343 
344 	printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
345 	show_registers(regs);
346 
347 	if (panic_on_io_nmi)
348 		panic("NMI IOCK error: Not continuing");
349 
350 	/* Re-enable the IOCK line, wait for a few seconds */
351 	reason = (reason & 0xf) | 8;
352 	outb(reason, 0x61);
353 
354 	i = 2000;
355 	while (--i)
356 		udelay(1000);
357 
358 	reason &= ~8;
359 	outb(reason, 0x61);
360 }
361 
362 static notrace __kprobes void
363 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
364 {
365 	if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
366 			NOTIFY_STOP)
367 		return;
368 #ifdef CONFIG_MCA
369 	/*
370 	 * Might actually be able to figure out what the guilty party
371 	 * is:
372 	 */
373 	if (MCA_bus) {
374 		mca_handle_nmi();
375 		return;
376 	}
377 #endif
378 	printk(KERN_EMERG
379 		"Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
380 			reason, smp_processor_id());
381 
382 	printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
383 	if (unknown_nmi_panic || panic_on_unrecovered_nmi)
384 		panic("NMI: Not continuing");
385 
386 	printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
387 }
388 
389 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
390 {
391 	unsigned char reason = 0;
392 	int cpu;
393 
394 	cpu = smp_processor_id();
395 
396 	/* Only the BSP gets external NMIs from the system. */
397 	if (!cpu)
398 		reason = get_nmi_reason();
399 
400 	if (!(reason & 0xc0)) {
401 		if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
402 								== NOTIFY_STOP)
403 			return;
404 
405 #ifdef CONFIG_X86_LOCAL_APIC
406 		if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT)
407 							== NOTIFY_STOP)
408 			return;
409 #endif
410 		unknown_nmi_error(reason, regs);
411 
412 		return;
413 	}
414 	if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
415 		return;
416 
417 	/* AK: following checks seem to be broken on modern chipsets. FIXME */
418 	if (reason & 0x80)
419 		mem_parity_error(reason, regs);
420 	if (reason & 0x40)
421 		io_check_error(reason, regs);
422 #ifdef CONFIG_X86_32
423 	/*
424 	 * Reassert NMI in case it became active meanwhile
425 	 * as it's edge-triggered:
426 	 */
427 	reassert_nmi();
428 #endif
429 }
430 
431 dotraplinkage notrace __kprobes void
432 do_nmi(struct pt_regs *regs, long error_code)
433 {
434 	nmi_enter();
435 
436 	inc_irq_stat(__nmi_count);
437 
438 	if (!ignore_nmis)
439 		default_do_nmi(regs);
440 
441 	nmi_exit();
442 }
443 
444 void stop_nmi(void)
445 {
446 	ignore_nmis++;
447 }
448 
449 void restart_nmi(void)
450 {
451 	ignore_nmis--;
452 }
453 
454 /* May run on IST stack. */
455 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
456 {
457 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
458 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
459 			== NOTIFY_STOP)
460 		return;
461 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
462 #ifdef CONFIG_KPROBES
463 	if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
464 			== NOTIFY_STOP)
465 		return;
466 #else
467 	if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
468 			== NOTIFY_STOP)
469 		return;
470 #endif
471 
472 	preempt_conditional_sti(regs);
473 	do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
474 	preempt_conditional_cli(regs);
475 }
476 
477 #ifdef CONFIG_X86_64
478 /*
479  * Help handler running on IST stack to switch back to user stack
480  * for scheduling or signal handling. The actual stack switch is done in
481  * entry.S
482  */
483 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
484 {
485 	struct pt_regs *regs = eregs;
486 	/* Did already sync */
487 	if (eregs == (struct pt_regs *)eregs->sp)
488 		;
489 	/* Exception from user space */
490 	else if (user_mode(eregs))
491 		regs = task_pt_regs(current);
492 	/*
493 	 * Exception from kernel and interrupts are enabled. Move to
494 	 * kernel process stack.
495 	 */
496 	else if (eregs->flags & X86_EFLAGS_IF)
497 		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
498 	if (eregs != regs)
499 		*regs = *eregs;
500 	return regs;
501 }
502 #endif
503 
504 /*
505  * Our handling of the processor debug registers is non-trivial.
506  * We do not clear them on entry and exit from the kernel. Therefore
507  * it is possible to get a watchpoint trap here from inside the kernel.
508  * However, the code in ./ptrace.c has ensured that the user can
509  * only set watchpoints on userspace addresses. Therefore the in-kernel
510  * watchpoint trap can only occur in code which is reading/writing
511  * from user space. Such code must not hold kernel locks (since it
512  * can equally take a page fault), therefore it is safe to call
513  * force_sig_info even though that claims and releases locks.
514  *
515  * Code in ./signal.c ensures that the debug control register
516  * is restored before we deliver any signal, and therefore that
517  * user code runs with the correct debug control register even though
518  * we clear it here.
519  *
520  * Being careful here means that we don't have to be as careful in a
521  * lot of more complicated places (task switching can be a bit lazy
522  * about restoring all the debug state, and ptrace doesn't have to
523  * find every occurrence of the TF bit that could be saved away even
524  * by user code)
525  *
526  * May run on IST stack.
527  */
528 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
529 {
530 	struct task_struct *tsk = current;
531 	int user_icebp = 0;
532 	unsigned long dr6;
533 	int si_code;
534 
535 	get_debugreg(dr6, 6);
536 
537 	/* Filter out all the reserved bits which are preset to 1 */
538 	dr6 &= ~DR6_RESERVED;
539 
540 	/*
541 	 * If dr6 has no reason to give us about the origin of this trap,
542 	 * then it's very likely the result of an icebp/int01 trap.
543 	 * User wants a sigtrap for that.
544 	 */
545 	if (!dr6 && user_mode(regs))
546 		user_icebp = 1;
547 
548 	/* Catch kmemcheck conditions first of all! */
549 	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
550 		return;
551 
552 	/* DR6 may or may not be cleared by the CPU */
553 	set_debugreg(0, 6);
554 
555 	/*
556 	 * The processor cleared BTF, so don't mark that we need it set.
557 	 */
558 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
559 
560 	/* Store the virtualized DR6 value */
561 	tsk->thread.debugreg6 = dr6;
562 
563 	if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
564 							SIGTRAP) == NOTIFY_STOP)
565 		return;
566 
567 	/* It's safe to allow irq's after DR6 has been saved */
568 	preempt_conditional_sti(regs);
569 
570 	if (regs->flags & X86_VM_MASK) {
571 		handle_vm86_trap((struct kernel_vm86_regs *) regs,
572 				error_code, 1);
573 		preempt_conditional_cli(regs);
574 		return;
575 	}
576 
577 	/*
578 	 * Single-stepping through system calls: ignore any exceptions in
579 	 * kernel space, but re-enable TF when returning to user mode.
580 	 *
581 	 * We already checked v86 mode above, so we can check for kernel mode
582 	 * by just checking the CPL of CS.
583 	 */
584 	if ((dr6 & DR_STEP) && !user_mode(regs)) {
585 		tsk->thread.debugreg6 &= ~DR_STEP;
586 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
587 		regs->flags &= ~X86_EFLAGS_TF;
588 	}
589 	si_code = get_si_code(tsk->thread.debugreg6);
590 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
591 		send_sigtrap(tsk, regs, error_code, si_code);
592 	preempt_conditional_cli(regs);
593 
594 	return;
595 }
596 
597 /*
598  * Note that we play around with the 'TS' bit in an attempt to get
599  * the correct behaviour even in the presence of the asynchronous
600  * IRQ13 behaviour
601  */
602 void math_error(struct pt_regs *regs, int error_code, int trapnr)
603 {
604 	struct task_struct *task = current;
605 	siginfo_t info;
606 	unsigned short err;
607 	char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
608 
609 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
610 		return;
611 	conditional_sti(regs);
612 
613 	if (!user_mode_vm(regs))
614 	{
615 		if (!fixup_exception(regs)) {
616 			task->thread.error_code = error_code;
617 			task->thread.trap_no = trapnr;
618 			die(str, regs, error_code);
619 		}
620 		return;
621 	}
622 
623 	/*
624 	 * Save the info for the exception handler and clear the error.
625 	 */
626 	save_init_fpu(task);
627 	task->thread.trap_no = trapnr;
628 	task->thread.error_code = error_code;
629 	info.si_signo = SIGFPE;
630 	info.si_errno = 0;
631 	info.si_addr = (void __user *)regs->ip;
632 	if (trapnr == 16) {
633 		unsigned short cwd, swd;
634 		/*
635 		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
636 		 * status.  0x3f is the exception bits in these regs, 0x200 is the
637 		 * C1 reg you need in case of a stack fault, 0x040 is the stack
638 		 * fault bit.  We should only be taking one exception at a time,
639 		 * so if this combination doesn't produce any single exception,
640 		 * then we have a bad program that isn't synchronizing its FPU usage
641 		 * and it will suffer the consequences since we won't be able to
642 		 * fully reproduce the context of the exception
643 		 */
644 		cwd = get_fpu_cwd(task);
645 		swd = get_fpu_swd(task);
646 
647 		err = swd & ~cwd;
648 	} else {
649 		/*
650 		 * The SIMD FPU exceptions are handled a little differently, as there
651 		 * is only a single status/control register.  Thus, to determine which
652 		 * unmasked exception was caught we must mask the exception mask bits
653 		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
654 		 */
655 		unsigned short mxcsr = get_fpu_mxcsr(task);
656 		err = ~(mxcsr >> 7) & mxcsr;
657 	}
658 
659 	if (err & 0x001) {	/* Invalid op */
660 		/*
661 		 * swd & 0x240 == 0x040: Stack Underflow
662 		 * swd & 0x240 == 0x240: Stack Overflow
663 		 * User must clear the SF bit (0x40) if set
664 		 */
665 		info.si_code = FPE_FLTINV;
666 	} else if (err & 0x004) { /* Divide by Zero */
667 		info.si_code = FPE_FLTDIV;
668 	} else if (err & 0x008) { /* Overflow */
669 		info.si_code = FPE_FLTOVF;
670 	} else if (err & 0x012) { /* Denormal, Underflow */
671 		info.si_code = FPE_FLTUND;
672 	} else if (err & 0x020) { /* Precision */
673 		info.si_code = FPE_FLTRES;
674 	} else {
675 		/*
676 		 * If we're using IRQ 13, or supposedly even some trap 16
677 		 * implementations, it's possible we get a spurious trap...
678 		 */
679 		return;		/* Spurious trap, no error */
680 	}
681 	force_sig_info(SIGFPE, &info, task);
682 }
683 
684 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
685 {
686 #ifdef CONFIG_X86_32
687 	ignore_fpu_irq = 1;
688 #endif
689 
690 	math_error(regs, error_code, 16);
691 }
692 
693 dotraplinkage void
694 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
695 {
696 	math_error(regs, error_code, 19);
697 }
698 
699 dotraplinkage void
700 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
701 {
702 	conditional_sti(regs);
703 #if 0
704 	/* No need to warn about this any longer. */
705 	printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
706 #endif
707 }
708 
709 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
710 {
711 }
712 
713 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
714 {
715 }
716 
717 /*
718  * __math_state_restore assumes that cr0.TS is already clear and the
719  * fpu state is all ready for use.  Used during context switch.
720  */
721 void __math_state_restore(void)
722 {
723 	struct thread_info *thread = current_thread_info();
724 	struct task_struct *tsk = thread->task;
725 
726 	/*
727 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
728 	 */
729 	if (unlikely(restore_fpu_checking(tsk))) {
730 		stts();
731 		force_sig(SIGSEGV, tsk);
732 		return;
733 	}
734 
735 	thread->status |= TS_USEDFPU;	/* So we fnsave on switch_to() */
736 	tsk->fpu_counter++;
737 }
738 
739 /*
740  * 'math_state_restore()' saves the current math information in the
741  * old math state array, and gets the new ones from the current task
742  *
743  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
744  * Don't touch unless you *really* know how it works.
745  *
746  * Must be called with kernel preemption disabled (in this case,
747  * local interrupts are disabled at the call-site in entry.S).
748  */
749 asmlinkage void math_state_restore(void)
750 {
751 	struct thread_info *thread = current_thread_info();
752 	struct task_struct *tsk = thread->task;
753 
754 	if (!tsk_used_math(tsk)) {
755 		local_irq_enable();
756 		/*
757 		 * does a slab alloc which can sleep
758 		 */
759 		if (init_fpu(tsk)) {
760 			/*
761 			 * ran out of memory!
762 			 */
763 			do_group_exit(SIGKILL);
764 			return;
765 		}
766 		local_irq_disable();
767 	}
768 
769 	clts();				/* Allow maths ops (or we recurse) */
770 
771 	__math_state_restore();
772 }
773 EXPORT_SYMBOL_GPL(math_state_restore);
774 
775 dotraplinkage void __kprobes
776 do_device_not_available(struct pt_regs *regs, long error_code)
777 {
778 #ifdef CONFIG_MATH_EMULATION
779 	if (read_cr0() & X86_CR0_EM) {
780 		struct math_emu_info info = { };
781 
782 		conditional_sti(regs);
783 
784 		info.regs = regs;
785 		math_emulate(&info);
786 		return;
787 	}
788 #endif
789 	math_state_restore(); /* interrupts still off */
790 #ifdef CONFIG_X86_32
791 	conditional_sti(regs);
792 #endif
793 }
794 
795 #ifdef CONFIG_X86_32
796 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
797 {
798 	siginfo_t info;
799 	local_irq_enable();
800 
801 	info.si_signo = SIGILL;
802 	info.si_errno = 0;
803 	info.si_code = ILL_BADSTK;
804 	info.si_addr = NULL;
805 	if (notify_die(DIE_TRAP, "iret exception",
806 			regs, error_code, 32, SIGILL) == NOTIFY_STOP)
807 		return;
808 	do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
809 }
810 #endif
811 
812 /* Set of traps needed for early debugging. */
813 void __init early_trap_init(void)
814 {
815 	set_intr_gate_ist(1, &debug, DEBUG_STACK);
816 	/* int3 can be called from all */
817 	set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
818 	set_intr_gate(14, &page_fault);
819 	load_idt(&idt_descr);
820 }
821 
822 void __init trap_init(void)
823 {
824 	int i;
825 
826 #ifdef CONFIG_EISA
827 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
828 
829 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
830 		EISA_bus = 1;
831 	early_iounmap(p, 4);
832 #endif
833 
834 	set_intr_gate(0, &divide_error);
835 	set_intr_gate_ist(2, &nmi, NMI_STACK);
836 	/* int4 can be called from all */
837 	set_system_intr_gate(4, &overflow);
838 	set_intr_gate(5, &bounds);
839 	set_intr_gate(6, &invalid_op);
840 	set_intr_gate(7, &device_not_available);
841 #ifdef CONFIG_X86_32
842 	set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
843 #else
844 	set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
845 #endif
846 	set_intr_gate(9, &coprocessor_segment_overrun);
847 	set_intr_gate(10, &invalid_TSS);
848 	set_intr_gate(11, &segment_not_present);
849 	set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
850 	set_intr_gate(13, &general_protection);
851 	set_intr_gate(15, &spurious_interrupt_bug);
852 	set_intr_gate(16, &coprocessor_error);
853 	set_intr_gate(17, &alignment_check);
854 #ifdef CONFIG_X86_MCE
855 	set_intr_gate_ist(18, &machine_check, MCE_STACK);
856 #endif
857 	set_intr_gate(19, &simd_coprocessor_error);
858 
859 	/* Reserve all the builtin and the syscall vector: */
860 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
861 		set_bit(i, used_vectors);
862 
863 #ifdef CONFIG_IA32_EMULATION
864 	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
865 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
866 #endif
867 
868 #ifdef CONFIG_X86_32
869 	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
870 	set_bit(SYSCALL_VECTOR, used_vectors);
871 #endif
872 
873 	/*
874 	 * Should be a barrier for any external CPU state:
875 	 */
876 	cpu_init();
877 
878 	x86_init.irqs.trap_init();
879 }
880