xref: /linux/arch/x86/kernel/traps.c (revision 913df4453f85f1fe79b35ecf3c9a0c0b707d22a2)
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/kernel.h>
19 #include <linux/module.h>
20 #include <linux/ptrace.h>
21 #include <linux/string.h>
22 #include <linux/delay.h>
23 #include <linux/errno.h>
24 #include <linux/kexec.h>
25 #include <linux/sched.h>
26 #include <linux/timer.h>
27 #include <linux/init.h>
28 #include <linux/bug.h>
29 #include <linux/nmi.h>
30 #include <linux/mm.h>
31 #include <linux/smp.h>
32 #include <linux/io.h>
33 
34 #ifdef CONFIG_EISA
35 #include <linux/ioport.h>
36 #include <linux/eisa.h>
37 #endif
38 
39 #ifdef CONFIG_MCA
40 #include <linux/mca.h>
41 #endif
42 
43 #if defined(CONFIG_EDAC)
44 #include <linux/edac.h>
45 #endif
46 
47 #include <asm/kmemcheck.h>
48 #include <asm/stacktrace.h>
49 #include <asm/processor.h>
50 #include <asm/debugreg.h>
51 #include <asm/atomic.h>
52 #include <asm/system.h>
53 #include <asm/traps.h>
54 #include <asm/desc.h>
55 #include <asm/i387.h>
56 #include <asm/mce.h>
57 
58 #include <asm/mach_traps.h>
59 
60 #ifdef CONFIG_X86_64
61 #include <asm/x86_init.h>
62 #include <asm/pgalloc.h>
63 #include <asm/proto.h>
64 #else
65 #include <asm/processor-flags.h>
66 #include <asm/setup.h>
67 
68 asmlinkage int system_call(void);
69 
70 /* Do we ignore FPU interrupts ? */
71 char ignore_fpu_irq;
72 
73 /*
74  * The IDT has to be page-aligned to simplify the Pentium
75  * F0 0F bug workaround.
76  */
77 gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
78 #endif
79 
80 DECLARE_BITMAP(used_vectors, NR_VECTORS);
81 EXPORT_SYMBOL_GPL(used_vectors);
82 
83 static int ignore_nmis;
84 
85 static inline void conditional_sti(struct pt_regs *regs)
86 {
87 	if (regs->flags & X86_EFLAGS_IF)
88 		local_irq_enable();
89 }
90 
91 static inline void preempt_conditional_sti(struct pt_regs *regs)
92 {
93 	inc_preempt_count();
94 	if (regs->flags & X86_EFLAGS_IF)
95 		local_irq_enable();
96 }
97 
98 static inline void conditional_cli(struct pt_regs *regs)
99 {
100 	if (regs->flags & X86_EFLAGS_IF)
101 		local_irq_disable();
102 }
103 
104 static inline void preempt_conditional_cli(struct pt_regs *regs)
105 {
106 	if (regs->flags & X86_EFLAGS_IF)
107 		local_irq_disable();
108 	dec_preempt_count();
109 }
110 
111 #ifdef CONFIG_X86_32
112 static inline void
113 die_if_kernel(const char *str, struct pt_regs *regs, long err)
114 {
115 	if (!user_mode_vm(regs))
116 		die(str, regs, err);
117 }
118 #endif
119 
120 static void __kprobes
121 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
122 	long error_code, siginfo_t *info)
123 {
124 	struct task_struct *tsk = current;
125 
126 #ifdef CONFIG_X86_32
127 	if (regs->flags & X86_VM_MASK) {
128 		/*
129 		 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
130 		 * On nmi (interrupt 2), do_trap should not be called.
131 		 */
132 		if (trapnr < 6)
133 			goto vm86_trap;
134 		goto trap_signal;
135 	}
136 #endif
137 
138 	if (!user_mode(regs))
139 		goto kernel_trap;
140 
141 #ifdef CONFIG_X86_32
142 trap_signal:
143 #endif
144 	/*
145 	 * We want error_code and trap_no set for userspace faults and
146 	 * kernelspace faults which result in die(), but not
147 	 * kernelspace faults which are fixed up.  die() gives the
148 	 * process no chance to handle the signal and notice the
149 	 * kernel fault information, so that won't result in polluting
150 	 * the information about previously queued, but not yet
151 	 * delivered, faults.  See also do_general_protection below.
152 	 */
153 	tsk->thread.error_code = error_code;
154 	tsk->thread.trap_no = trapnr;
155 
156 #ifdef CONFIG_X86_64
157 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
158 	    printk_ratelimit()) {
159 		printk(KERN_INFO
160 		       "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
161 		       tsk->comm, tsk->pid, str,
162 		       regs->ip, regs->sp, error_code);
163 		print_vma_addr(" in ", regs->ip);
164 		printk("\n");
165 	}
166 #endif
167 
168 	if (info)
169 		force_sig_info(signr, info, tsk);
170 	else
171 		force_sig(signr, tsk);
172 	return;
173 
174 kernel_trap:
175 	if (!fixup_exception(regs)) {
176 		tsk->thread.error_code = error_code;
177 		tsk->thread.trap_no = trapnr;
178 		die(str, regs, error_code);
179 	}
180 	return;
181 
182 #ifdef CONFIG_X86_32
183 vm86_trap:
184 	if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
185 						error_code, trapnr))
186 		goto trap_signal;
187 	return;
188 #endif
189 }
190 
191 #define DO_ERROR(trapnr, signr, str, name)				\
192 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
193 {									\
194 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
195 							== NOTIFY_STOP)	\
196 		return;							\
197 	conditional_sti(regs);						\
198 	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
199 }
200 
201 #define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)		\
202 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
203 {									\
204 	siginfo_t info;							\
205 	info.si_signo = signr;						\
206 	info.si_errno = 0;						\
207 	info.si_code = sicode;						\
208 	info.si_addr = (void __user *)siaddr;				\
209 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
210 							== NOTIFY_STOP)	\
211 		return;							\
212 	conditional_sti(regs);						\
213 	do_trap(trapnr, signr, str, regs, error_code, &info);		\
214 }
215 
216 DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
217 DO_ERROR(4, SIGSEGV, "overflow", overflow)
218 DO_ERROR(5, SIGSEGV, "bounds", bounds)
219 DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
220 DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
221 DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
222 DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
223 #ifdef CONFIG_X86_32
224 DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
225 #endif
226 DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
227 
228 #ifdef CONFIG_X86_64
229 /* Runs on IST stack */
230 dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
231 {
232 	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
233 			12, SIGBUS) == NOTIFY_STOP)
234 		return;
235 	preempt_conditional_sti(regs);
236 	do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
237 	preempt_conditional_cli(regs);
238 }
239 
240 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
241 {
242 	static const char str[] = "double fault";
243 	struct task_struct *tsk = current;
244 
245 	/* Return not checked because double check cannot be ignored */
246 	notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
247 
248 	tsk->thread.error_code = error_code;
249 	tsk->thread.trap_no = 8;
250 
251 	/*
252 	 * This is always a kernel trap and never fixable (and thus must
253 	 * never return).
254 	 */
255 	for (;;)
256 		die(str, regs, error_code);
257 }
258 #endif
259 
260 dotraplinkage void __kprobes
261 do_general_protection(struct pt_regs *regs, long error_code)
262 {
263 	struct task_struct *tsk;
264 
265 	conditional_sti(regs);
266 
267 #ifdef CONFIG_X86_32
268 	if (regs->flags & X86_VM_MASK)
269 		goto gp_in_vm86;
270 #endif
271 
272 	tsk = current;
273 	if (!user_mode(regs))
274 		goto gp_in_kernel;
275 
276 	tsk->thread.error_code = error_code;
277 	tsk->thread.trap_no = 13;
278 
279 	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
280 			printk_ratelimit()) {
281 		printk(KERN_INFO
282 			"%s[%d] general protection ip:%lx sp:%lx error:%lx",
283 			tsk->comm, task_pid_nr(tsk),
284 			regs->ip, regs->sp, error_code);
285 		print_vma_addr(" in ", regs->ip);
286 		printk("\n");
287 	}
288 
289 	force_sig(SIGSEGV, tsk);
290 	return;
291 
292 #ifdef CONFIG_X86_32
293 gp_in_vm86:
294 	local_irq_enable();
295 	handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
296 	return;
297 #endif
298 
299 gp_in_kernel:
300 	if (fixup_exception(regs))
301 		return;
302 
303 	tsk->thread.error_code = error_code;
304 	tsk->thread.trap_no = 13;
305 	if (notify_die(DIE_GPF, "general protection fault", regs,
306 				error_code, 13, SIGSEGV) == NOTIFY_STOP)
307 		return;
308 	die("general protection fault", regs, error_code);
309 }
310 
311 static notrace __kprobes void
312 mem_parity_error(unsigned char reason, struct pt_regs *regs)
313 {
314 	printk(KERN_EMERG
315 		"Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
316 			reason, smp_processor_id());
317 
318 	printk(KERN_EMERG
319 		"You have some hardware problem, likely on the PCI bus.\n");
320 
321 #if defined(CONFIG_EDAC)
322 	if (edac_handler_set()) {
323 		edac_atomic_assert_error();
324 		return;
325 	}
326 #endif
327 
328 	if (panic_on_unrecovered_nmi)
329 		panic("NMI: Not continuing");
330 
331 	printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
332 
333 	/* Clear and disable the memory parity error line. */
334 	reason = (reason & 0xf) | 4;
335 	outb(reason, 0x61);
336 }
337 
338 static notrace __kprobes void
339 io_check_error(unsigned char reason, struct pt_regs *regs)
340 {
341 	unsigned long i;
342 
343 	printk(KERN_EMERG "NMI: IOCK error (debug interrupt?)\n");
344 	show_registers(regs);
345 
346 	if (panic_on_io_nmi)
347 		panic("NMI IOCK error: Not continuing");
348 
349 	/* Re-enable the IOCK line, wait for a few seconds */
350 	reason = (reason & 0xf) | 8;
351 	outb(reason, 0x61);
352 
353 	i = 2000;
354 	while (--i)
355 		udelay(1000);
356 
357 	reason &= ~8;
358 	outb(reason, 0x61);
359 }
360 
361 static notrace __kprobes void
362 unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
363 {
364 	if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
365 			NOTIFY_STOP)
366 		return;
367 #ifdef CONFIG_MCA
368 	/*
369 	 * Might actually be able to figure out what the guilty party
370 	 * is:
371 	 */
372 	if (MCA_bus) {
373 		mca_handle_nmi();
374 		return;
375 	}
376 #endif
377 	printk(KERN_EMERG
378 		"Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
379 			reason, smp_processor_id());
380 
381 	printk(KERN_EMERG "Do you have a strange power saving mode enabled?\n");
382 	if (panic_on_unrecovered_nmi)
383 		panic("NMI: Not continuing");
384 
385 	printk(KERN_EMERG "Dazed and confused, but trying to continue\n");
386 }
387 
388 static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
389 {
390 	unsigned char reason = 0;
391 	int cpu;
392 
393 	cpu = smp_processor_id();
394 
395 	/* Only the BSP gets external NMIs from the system. */
396 	if (!cpu)
397 		reason = get_nmi_reason();
398 
399 	if (!(reason & 0xc0)) {
400 		if (notify_die(DIE_NMI_IPI, "nmi_ipi", regs, reason, 2, SIGINT)
401 								== NOTIFY_STOP)
402 			return;
403 #ifdef CONFIG_X86_LOCAL_APIC
404 		/*
405 		 * Ok, so this is none of the documented NMI sources,
406 		 * so it must be the NMI watchdog.
407 		 */
408 		if (nmi_watchdog_tick(regs, reason))
409 			return;
410 		if (!do_nmi_callback(regs, cpu))
411 			unknown_nmi_error(reason, regs);
412 #else
413 		unknown_nmi_error(reason, regs);
414 #endif
415 
416 		return;
417 	}
418 	if (notify_die(DIE_NMI, "nmi", regs, reason, 2, SIGINT) == NOTIFY_STOP)
419 		return;
420 
421 	/* AK: following checks seem to be broken on modern chipsets. FIXME */
422 	if (reason & 0x80)
423 		mem_parity_error(reason, regs);
424 	if (reason & 0x40)
425 		io_check_error(reason, regs);
426 #ifdef CONFIG_X86_32
427 	/*
428 	 * Reassert NMI in case it became active meanwhile
429 	 * as it's edge-triggered:
430 	 */
431 	reassert_nmi();
432 #endif
433 }
434 
435 dotraplinkage notrace __kprobes void
436 do_nmi(struct pt_regs *regs, long error_code)
437 {
438 	nmi_enter();
439 
440 	inc_irq_stat(__nmi_count);
441 
442 	if (!ignore_nmis)
443 		default_do_nmi(regs);
444 
445 	nmi_exit();
446 }
447 
448 void stop_nmi(void)
449 {
450 	acpi_nmi_disable();
451 	ignore_nmis++;
452 }
453 
454 void restart_nmi(void)
455 {
456 	ignore_nmis--;
457 	acpi_nmi_enable();
458 }
459 
460 /* May run on IST stack. */
461 dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
462 {
463 #ifdef CONFIG_KPROBES
464 	if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
465 			== NOTIFY_STOP)
466 		return;
467 #else
468 	if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
469 			== NOTIFY_STOP)
470 		return;
471 #endif
472 
473 	preempt_conditional_sti(regs);
474 	do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
475 	preempt_conditional_cli(regs);
476 }
477 
478 #ifdef CONFIG_X86_64
479 /*
480  * Help handler running on IST stack to switch back to user stack
481  * for scheduling or signal handling. The actual stack switch is done in
482  * entry.S
483  */
484 asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
485 {
486 	struct pt_regs *regs = eregs;
487 	/* Did already sync */
488 	if (eregs == (struct pt_regs *)eregs->sp)
489 		;
490 	/* Exception from user space */
491 	else if (user_mode(eregs))
492 		regs = task_pt_regs(current);
493 	/*
494 	 * Exception from kernel and interrupts are enabled. Move to
495 	 * kernel process stack.
496 	 */
497 	else if (eregs->flags & X86_EFLAGS_IF)
498 		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
499 	if (eregs != regs)
500 		*regs = *eregs;
501 	return regs;
502 }
503 #endif
504 
505 /*
506  * Our handling of the processor debug registers is non-trivial.
507  * We do not clear them on entry and exit from the kernel. Therefore
508  * it is possible to get a watchpoint trap here from inside the kernel.
509  * However, the code in ./ptrace.c has ensured that the user can
510  * only set watchpoints on userspace addresses. Therefore the in-kernel
511  * watchpoint trap can only occur in code which is reading/writing
512  * from user space. Such code must not hold kernel locks (since it
513  * can equally take a page fault), therefore it is safe to call
514  * force_sig_info even though that claims and releases locks.
515  *
516  * Code in ./signal.c ensures that the debug control register
517  * is restored before we deliver any signal, and therefore that
518  * user code runs with the correct debug control register even though
519  * we clear it here.
520  *
521  * Being careful here means that we don't have to be as careful in a
522  * lot of more complicated places (task switching can be a bit lazy
523  * about restoring all the debug state, and ptrace doesn't have to
524  * find every occurrence of the TF bit that could be saved away even
525  * by user code)
526  *
527  * May run on IST stack.
528  */
529 dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
530 {
531 	struct task_struct *tsk = current;
532 	unsigned long condition;
533 	int si_code;
534 
535 	get_debugreg(condition, 6);
536 
537 	/* Catch kmemcheck conditions first of all! */
538 	if (condition & DR_STEP && kmemcheck_trap(regs))
539 		return;
540 
541 	/*
542 	 * The processor cleared BTF, so don't mark that we need it set.
543 	 */
544 	clear_tsk_thread_flag(tsk, TIF_DEBUGCTLMSR);
545 	tsk->thread.debugctlmsr = 0;
546 
547 	if (notify_die(DIE_DEBUG, "debug", regs, condition, error_code,
548 						SIGTRAP) == NOTIFY_STOP)
549 		return;
550 
551 	/* It's safe to allow irq's after DR6 has been saved */
552 	preempt_conditional_sti(regs);
553 
554 	/* Mask out spurious debug traps due to lazy DR7 setting */
555 	if (condition & (DR_TRAP0|DR_TRAP1|DR_TRAP2|DR_TRAP3)) {
556 		if (!tsk->thread.debugreg7)
557 			goto clear_dr7;
558 	}
559 
560 #ifdef CONFIG_X86_32
561 	if (regs->flags & X86_VM_MASK)
562 		goto debug_vm86;
563 #endif
564 
565 	/* Save debug status register where ptrace can see it */
566 	tsk->thread.debugreg6 = condition;
567 
568 	/*
569 	 * Single-stepping through TF: make sure we ignore any events in
570 	 * kernel space (but re-enable TF when returning to user mode).
571 	 */
572 	if (condition & DR_STEP) {
573 		if (!user_mode(regs))
574 			goto clear_TF_reenable;
575 	}
576 
577 	si_code = get_si_code(condition);
578 	/* Ok, finally something we can handle */
579 	send_sigtrap(tsk, regs, error_code, si_code);
580 
581 	/*
582 	 * Disable additional traps. They'll be re-enabled when
583 	 * the signal is delivered.
584 	 */
585 clear_dr7:
586 	set_debugreg(0, 7);
587 	preempt_conditional_cli(regs);
588 	return;
589 
590 #ifdef CONFIG_X86_32
591 debug_vm86:
592 	/* reenable preemption: handle_vm86_trap() might sleep */
593 	dec_preempt_count();
594 	handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code, 1);
595 	conditional_cli(regs);
596 	return;
597 #endif
598 
599 clear_TF_reenable:
600 	set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
601 	regs->flags &= ~X86_EFLAGS_TF;
602 	preempt_conditional_cli(regs);
603 	return;
604 }
605 
606 #ifdef CONFIG_X86_64
607 static int kernel_math_error(struct pt_regs *regs, const char *str, int trapnr)
608 {
609 	if (fixup_exception(regs))
610 		return 1;
611 
612 	notify_die(DIE_GPF, str, regs, 0, trapnr, SIGFPE);
613 	/* Illegal floating point operation in the kernel */
614 	current->thread.trap_no = trapnr;
615 	die(str, regs, 0);
616 	return 0;
617 }
618 #endif
619 
620 /*
621  * Note that we play around with the 'TS' bit in an attempt to get
622  * the correct behaviour even in the presence of the asynchronous
623  * IRQ13 behaviour
624  */
625 void math_error(void __user *ip)
626 {
627 	struct task_struct *task;
628 	siginfo_t info;
629 	unsigned short cwd, swd, err;
630 
631 	/*
632 	 * Save the info for the exception handler and clear the error.
633 	 */
634 	task = current;
635 	save_init_fpu(task);
636 	task->thread.trap_no = 16;
637 	task->thread.error_code = 0;
638 	info.si_signo = SIGFPE;
639 	info.si_errno = 0;
640 	info.si_addr = ip;
641 	/*
642 	 * (~cwd & swd) will mask out exceptions that are not set to unmasked
643 	 * status.  0x3f is the exception bits in these regs, 0x200 is the
644 	 * C1 reg you need in case of a stack fault, 0x040 is the stack
645 	 * fault bit.  We should only be taking one exception at a time,
646 	 * so if this combination doesn't produce any single exception,
647 	 * then we have a bad program that isn't synchronizing its FPU usage
648 	 * and it will suffer the consequences since we won't be able to
649 	 * fully reproduce the context of the exception
650 	 */
651 	cwd = get_fpu_cwd(task);
652 	swd = get_fpu_swd(task);
653 
654 	err = swd & ~cwd;
655 
656 	if (err & 0x001) {	/* Invalid op */
657 		/*
658 		 * swd & 0x240 == 0x040: Stack Underflow
659 		 * swd & 0x240 == 0x240: Stack Overflow
660 		 * User must clear the SF bit (0x40) if set
661 		 */
662 		info.si_code = FPE_FLTINV;
663 	} else if (err & 0x004) { /* Divide by Zero */
664 		info.si_code = FPE_FLTDIV;
665 	} else if (err & 0x008) { /* Overflow */
666 		info.si_code = FPE_FLTOVF;
667 	} else if (err & 0x012) { /* Denormal, Underflow */
668 		info.si_code = FPE_FLTUND;
669 	} else if (err & 0x020) { /* Precision */
670 		info.si_code = FPE_FLTRES;
671 	} else {
672 		/*
673 		 * If we're using IRQ 13, or supposedly even some trap 16
674 		 * implementations, it's possible we get a spurious trap...
675 		 */
676 		return;		/* Spurious trap, no error */
677 	}
678 	force_sig_info(SIGFPE, &info, task);
679 }
680 
681 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
682 {
683 	conditional_sti(regs);
684 
685 #ifdef CONFIG_X86_32
686 	ignore_fpu_irq = 1;
687 #else
688 	if (!user_mode(regs) &&
689 	    kernel_math_error(regs, "kernel x87 math error", 16))
690 		return;
691 #endif
692 
693 	math_error((void __user *)regs->ip);
694 }
695 
696 static void simd_math_error(void __user *ip)
697 {
698 	struct task_struct *task;
699 	siginfo_t info;
700 	unsigned short mxcsr;
701 
702 	/*
703 	 * Save the info for the exception handler and clear the error.
704 	 */
705 	task = current;
706 	save_init_fpu(task);
707 	task->thread.trap_no = 19;
708 	task->thread.error_code = 0;
709 	info.si_signo = SIGFPE;
710 	info.si_errno = 0;
711 	info.si_code = __SI_FAULT;
712 	info.si_addr = ip;
713 	/*
714 	 * The SIMD FPU exceptions are handled a little differently, as there
715 	 * is only a single status/control register.  Thus, to determine which
716 	 * unmasked exception was caught we must mask the exception mask bits
717 	 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
718 	 */
719 	mxcsr = get_fpu_mxcsr(task);
720 	switch (~((mxcsr & 0x1f80) >> 7) & (mxcsr & 0x3f)) {
721 	case 0x000:
722 	default:
723 		break;
724 	case 0x001: /* Invalid Op */
725 		info.si_code = FPE_FLTINV;
726 		break;
727 	case 0x002: /* Denormalize */
728 	case 0x010: /* Underflow */
729 		info.si_code = FPE_FLTUND;
730 		break;
731 	case 0x004: /* Zero Divide */
732 		info.si_code = FPE_FLTDIV;
733 		break;
734 	case 0x008: /* Overflow */
735 		info.si_code = FPE_FLTOVF;
736 		break;
737 	case 0x020: /* Precision */
738 		info.si_code = FPE_FLTRES;
739 		break;
740 	}
741 	force_sig_info(SIGFPE, &info, task);
742 }
743 
744 dotraplinkage void
745 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
746 {
747 	conditional_sti(regs);
748 
749 #ifdef CONFIG_X86_32
750 	if (cpu_has_xmm) {
751 		/* Handle SIMD FPU exceptions on PIII+ processors. */
752 		ignore_fpu_irq = 1;
753 		simd_math_error((void __user *)regs->ip);
754 		return;
755 	}
756 	/*
757 	 * Handle strange cache flush from user space exception
758 	 * in all other cases.  This is undocumented behaviour.
759 	 */
760 	if (regs->flags & X86_VM_MASK) {
761 		handle_vm86_fault((struct kernel_vm86_regs *)regs, error_code);
762 		return;
763 	}
764 	current->thread.trap_no = 19;
765 	current->thread.error_code = error_code;
766 	die_if_kernel("cache flush denied", regs, error_code);
767 	force_sig(SIGSEGV, current);
768 #else
769 	if (!user_mode(regs) &&
770 			kernel_math_error(regs, "kernel simd math error", 19))
771 		return;
772 	simd_math_error((void __user *)regs->ip);
773 #endif
774 }
775 
776 dotraplinkage void
777 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
778 {
779 	conditional_sti(regs);
780 #if 0
781 	/* No need to warn about this any longer. */
782 	printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
783 #endif
784 }
785 
786 asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
787 {
788 }
789 
790 asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
791 {
792 }
793 
794 /*
795  * __math_state_restore assumes that cr0.TS is already clear and the
796  * fpu state is all ready for use.  Used during context switch.
797  */
798 void __math_state_restore(void)
799 {
800 	struct thread_info *thread = current_thread_info();
801 	struct task_struct *tsk = thread->task;
802 
803 	/*
804 	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
805 	 */
806 	if (unlikely(restore_fpu_checking(tsk))) {
807 		stts();
808 		force_sig(SIGSEGV, tsk);
809 		return;
810 	}
811 
812 	thread->status |= TS_USEDFPU;	/* So we fnsave on switch_to() */
813 	tsk->fpu_counter++;
814 }
815 
816 /*
817  * 'math_state_restore()' saves the current math information in the
818  * old math state array, and gets the new ones from the current task
819  *
820  * Careful.. There are problems with IBM-designed IRQ13 behaviour.
821  * Don't touch unless you *really* know how it works.
822  *
823  * Must be called with kernel preemption disabled (in this case,
824  * local interrupts are disabled at the call-site in entry.S).
825  */
826 asmlinkage void math_state_restore(void)
827 {
828 	struct thread_info *thread = current_thread_info();
829 	struct task_struct *tsk = thread->task;
830 
831 	if (!tsk_used_math(tsk)) {
832 		local_irq_enable();
833 		/*
834 		 * does a slab alloc which can sleep
835 		 */
836 		if (init_fpu(tsk)) {
837 			/*
838 			 * ran out of memory!
839 			 */
840 			do_group_exit(SIGKILL);
841 			return;
842 		}
843 		local_irq_disable();
844 	}
845 
846 	clts();				/* Allow maths ops (or we recurse) */
847 
848 	__math_state_restore();
849 }
850 EXPORT_SYMBOL_GPL(math_state_restore);
851 
852 #ifndef CONFIG_MATH_EMULATION
853 void math_emulate(struct math_emu_info *info)
854 {
855 	printk(KERN_EMERG
856 		"math-emulation not enabled and no coprocessor found.\n");
857 	printk(KERN_EMERG "killing %s.\n", current->comm);
858 	force_sig(SIGFPE, current);
859 	schedule();
860 }
861 #endif /* CONFIG_MATH_EMULATION */
862 
863 dotraplinkage void __kprobes
864 do_device_not_available(struct pt_regs *regs, long error_code)
865 {
866 #ifdef CONFIG_X86_32
867 	if (read_cr0() & X86_CR0_EM) {
868 		struct math_emu_info info = { };
869 
870 		conditional_sti(regs);
871 
872 		info.regs = regs;
873 		math_emulate(&info);
874 	} else {
875 		math_state_restore(); /* interrupts still off */
876 		conditional_sti(regs);
877 	}
878 #else
879 	math_state_restore();
880 #endif
881 }
882 
883 #ifdef CONFIG_X86_32
884 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
885 {
886 	siginfo_t info;
887 	local_irq_enable();
888 
889 	info.si_signo = SIGILL;
890 	info.si_errno = 0;
891 	info.si_code = ILL_BADSTK;
892 	info.si_addr = NULL;
893 	if (notify_die(DIE_TRAP, "iret exception",
894 			regs, error_code, 32, SIGILL) == NOTIFY_STOP)
895 		return;
896 	do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
897 }
898 #endif
899 
900 void __init trap_init(void)
901 {
902 	int i;
903 
904 #ifdef CONFIG_EISA
905 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
906 
907 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
908 		EISA_bus = 1;
909 	early_iounmap(p, 4);
910 #endif
911 
912 	set_intr_gate(0, &divide_error);
913 	set_intr_gate_ist(1, &debug, DEBUG_STACK);
914 	set_intr_gate_ist(2, &nmi, NMI_STACK);
915 	/* int3 can be called from all */
916 	set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
917 	/* int4 can be called from all */
918 	set_system_intr_gate(4, &overflow);
919 	set_intr_gate(5, &bounds);
920 	set_intr_gate(6, &invalid_op);
921 	set_intr_gate(7, &device_not_available);
922 #ifdef CONFIG_X86_32
923 	set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
924 #else
925 	set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
926 #endif
927 	set_intr_gate(9, &coprocessor_segment_overrun);
928 	set_intr_gate(10, &invalid_TSS);
929 	set_intr_gate(11, &segment_not_present);
930 	set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
931 	set_intr_gate(13, &general_protection);
932 	set_intr_gate(14, &page_fault);
933 	set_intr_gate(15, &spurious_interrupt_bug);
934 	set_intr_gate(16, &coprocessor_error);
935 	set_intr_gate(17, &alignment_check);
936 #ifdef CONFIG_X86_MCE
937 	set_intr_gate_ist(18, &machine_check, MCE_STACK);
938 #endif
939 	set_intr_gate(19, &simd_coprocessor_error);
940 
941 	/* Reserve all the builtin and the syscall vector: */
942 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
943 		set_bit(i, used_vectors);
944 
945 #ifdef CONFIG_IA32_EMULATION
946 	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
947 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
948 #endif
949 
950 #ifdef CONFIG_X86_32
951 	if (cpu_has_fxsr) {
952 		printk(KERN_INFO "Enabling fast FPU save and restore... ");
953 		set_in_cr4(X86_CR4_OSFXSR);
954 		printk("done.\n");
955 	}
956 	if (cpu_has_xmm) {
957 		printk(KERN_INFO
958 			"Enabling unmasked SIMD FPU exception support... ");
959 		set_in_cr4(X86_CR4_OSXMMEXCPT);
960 		printk("done.\n");
961 	}
962 
963 	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
964 	set_bit(SYSCALL_VECTOR, used_vectors);
965 #endif
966 
967 	/*
968 	 * Should be a barrier for any external CPU state:
969 	 */
970 	cpu_init();
971 
972 	x86_init.irqs.trap_init();
973 }
974