Linux Audio

Check our new training course

In-person Linux kernel drivers training

Jun 16-20, 2025
Register
Loading...
v3.1
  1/*
  2 *  Copyright (C) 1991, 1992  Linus Torvalds
  3 *  Copyright (C) 2000, 2001, 2002 Andi Kleen, SuSE Labs
  4 *
  5 *  Pentium III FXSR, SSE support
  6 *	Gareth Hughes <gareth@valinux.com>, May 2000
  7 */
  8
  9/*
 10 * Handle hardware traps and faults.
 11 */
 
 
 
 
 12#include <linux/interrupt.h>
 13#include <linux/kallsyms.h>
 14#include <linux/spinlock.h>
 15#include <linux/kprobes.h>
 16#include <linux/uaccess.h>
 17#include <linux/kdebug.h>
 18#include <linux/kgdb.h>
 19#include <linux/kernel.h>
 20#include <linux/module.h>
 21#include <linux/ptrace.h>
 22#include <linux/string.h>
 23#include <linux/delay.h>
 24#include <linux/errno.h>
 25#include <linux/kexec.h>
 26#include <linux/sched.h>
 27#include <linux/timer.h>
 28#include <linux/init.h>
 29#include <linux/bug.h>
 30#include <linux/nmi.h>
 31#include <linux/mm.h>
 32#include <linux/smp.h>
 33#include <linux/io.h>
 34
 35#ifdef CONFIG_EISA
 36#include <linux/ioport.h>
 37#include <linux/eisa.h>
 38#endif
 39
 40#ifdef CONFIG_MCA
 41#include <linux/mca.h>
 42#endif
 43
 44#if defined(CONFIG_EDAC)
 45#include <linux/edac.h>
 46#endif
 47
 48#include <asm/kmemcheck.h>
 49#include <asm/stacktrace.h>
 50#include <asm/processor.h>
 51#include <asm/debugreg.h>
 52#include <linux/atomic.h>
 53#include <asm/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
 69asmlinkage int system_call(void);
 70
 71/* Do we ignore FPU interrupts ? */
 72char ignore_fpu_irq;
 73
 74/*
 75 * The IDT has to be page-aligned to simplify the Pentium
 76 * F0 0F bug workaround.
 77 */
 78gate_desc idt_table[NR_VECTORS] __page_aligned_data = { { { { 0, 0 } } }, };
 79#endif
 80
 
 
 
 81DECLARE_BITMAP(used_vectors, NR_VECTORS);
 82EXPORT_SYMBOL_GPL(used_vectors);
 83
 84static int ignore_nmis;
 85
 86int unknown_nmi_panic;
 87/*
 88 * Prevent NMI reason port (0x61) being accessed simultaneously, can
 89 * only be used in NMI handler.
 90 */
 91static DEFINE_RAW_SPINLOCK(nmi_reason_lock);
 92
 93static inline void conditional_sti(struct pt_regs *regs)
 94{
 95	if (regs->flags & X86_EFLAGS_IF)
 96		local_irq_enable();
 97}
 98
 99static inline void preempt_conditional_sti(struct pt_regs *regs)
100{
101	inc_preempt_count();
102	if (regs->flags & X86_EFLAGS_IF)
103		local_irq_enable();
104}
105
106static inline void conditional_cli(struct pt_regs *regs)
107{
108	if (regs->flags & X86_EFLAGS_IF)
109		local_irq_disable();
110}
111
112static inline void preempt_conditional_cli(struct pt_regs *regs)
113{
114	if (regs->flags & X86_EFLAGS_IF)
115		local_irq_disable();
116	dec_preempt_count();
117}
118
119static void __kprobes
120do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
121	long error_code, siginfo_t *info)
122{
123	struct task_struct *tsk = current;
124
125#ifdef CONFIG_X86_32
126	if (regs->flags & X86_VM_MASK) {
127		/*
128		 * traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
129		 * On nmi (interrupt 2), do_trap should not be called.
130		 */
131		if (trapnr < 6)
132			goto vm86_trap;
133		goto trap_signal;
 
 
 
134	}
135#endif
 
 
 
 
 
 
 
 
136
137	if (!user_mode(regs))
138		goto kernel_trap;
139
140#ifdef CONFIG_X86_32
141trap_signal:
142#endif
 
 
 
 
 
 
143	/*
144	 * We want error_code and trap_no set for userspace faults and
145	 * kernelspace faults which result in die(), but not
146	 * kernelspace faults which are fixed up.  die() gives the
147	 * process no chance to handle the signal and notice the
148	 * kernel fault information, so that won't result in polluting
149	 * the information about previously queued, but not yet
150	 * delivered, faults.  See also do_general_protection below.
151	 */
152	tsk->thread.error_code = error_code;
153	tsk->thread.trap_no = trapnr;
154
155#ifdef CONFIG_X86_64
156	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
157	    printk_ratelimit()) {
158		printk(KERN_INFO
159		       "%s[%d] trap %s ip:%lx sp:%lx error:%lx",
160		       tsk->comm, tsk->pid, str,
161		       regs->ip, regs->sp, error_code);
162		print_vma_addr(" in ", regs->ip);
163		printk("\n");
164	}
165#endif
166
167	if (info)
168		force_sig_info(signr, info, tsk);
169	else
170		force_sig(signr, tsk);
171	return;
172
173kernel_trap:
174	if (!fixup_exception(regs)) {
175		tsk->thread.error_code = error_code;
176		tsk->thread.trap_no = trapnr;
177		die(str, regs, error_code);
178	}
179	return;
180
181#ifdef CONFIG_X86_32
182vm86_trap:
183	if (handle_vm86_trap((struct kernel_vm86_regs *) regs,
184						error_code, trapnr))
185		goto trap_signal;
186	return;
187#endif
188}
189
190#define DO_ERROR(trapnr, signr, str, name)				\
191dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
192{									\
193	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
194							== NOTIFY_STOP)	\
 
 
 
 
195		return;							\
 
196	conditional_sti(regs);						\
197	do_trap(trapnr, signr, str, regs, error_code, NULL);		\
 
198}
199
200#define DO_ERROR_INFO(trapnr, signr, str, name, sicode, siaddr)		\
201dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
202{									\
203	siginfo_t info;							\
 
 
204	info.si_signo = signr;						\
205	info.si_errno = 0;						\
206	info.si_code = sicode;						\
207	info.si_addr = (void __user *)siaddr;				\
208	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr)	\
209							== NOTIFY_STOP)	\
 
 
210		return;							\
 
211	conditional_sti(regs);						\
212	do_trap(trapnr, signr, str, regs, error_code, &info);		\
 
213}
214
215DO_ERROR_INFO(0, SIGFPE, "divide error", divide_error, FPE_INTDIV, regs->ip)
216DO_ERROR(4, SIGSEGV, "overflow", overflow)
217DO_ERROR(5, SIGSEGV, "bounds", bounds)
218DO_ERROR_INFO(6, SIGILL, "invalid opcode", invalid_op, ILL_ILLOPN, regs->ip)
219DO_ERROR(9, SIGFPE, "coprocessor segment overrun", coprocessor_segment_overrun)
220DO_ERROR(10, SIGSEGV, "invalid TSS", invalid_TSS)
221DO_ERROR(11, SIGBUS, "segment not present", segment_not_present)
222#ifdef CONFIG_X86_32
223DO_ERROR(12, SIGBUS, "stack segment", stack_segment)
224#endif
225DO_ERROR_INFO(17, SIGBUS, "alignment check", alignment_check, BUS_ADRALN, 0)
226
227#ifdef CONFIG_X86_64
228/* Runs on IST stack */
229dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
230{
 
 
 
231	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
232			12, SIGBUS) == NOTIFY_STOP)
233		return;
234	preempt_conditional_sti(regs);
235	do_trap(12, SIGBUS, "stack segment", regs, error_code, NULL);
236	preempt_conditional_cli(regs);
 
237}
238
239dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
240{
241	static const char str[] = "double fault";
242	struct task_struct *tsk = current;
243
 
244	/* Return not checked because double check cannot be ignored */
245	notify_die(DIE_TRAP, str, regs, error_code, 8, SIGSEGV);
246
247	tsk->thread.error_code = error_code;
248	tsk->thread.trap_no = 8;
249
 
 
 
250	/*
251	 * This is always a kernel trap and never fixable (and thus must
252	 * never return).
253	 */
254	for (;;)
255		die(str, regs, error_code);
256}
257#endif
258
259dotraplinkage void __kprobes
260do_general_protection(struct pt_regs *regs, long error_code)
261{
262	struct task_struct *tsk;
 
263
 
264	conditional_sti(regs);
265
266#ifdef CONFIG_X86_32
267	if (regs->flags & X86_VM_MASK)
268		goto gp_in_vm86;
 
 
 
269#endif
270
271	tsk = current;
272	if (!user_mode(regs))
273		goto gp_in_kernel;
 
 
 
 
 
 
 
 
 
274
275	tsk->thread.error_code = error_code;
276	tsk->thread.trap_no = 13;
277
278	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
279			printk_ratelimit()) {
280		printk(KERN_INFO
281			"%s[%d] general protection ip:%lx sp:%lx error:%lx",
282			tsk->comm, task_pid_nr(tsk),
283			regs->ip, regs->sp, error_code);
284		print_vma_addr(" in ", regs->ip);
285		printk("\n");
286	}
287
288	force_sig(SIGSEGV, tsk);
289	return;
290
291#ifdef CONFIG_X86_32
292gp_in_vm86:
293	local_irq_enable();
294	handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
295	return;
296#endif
297
298gp_in_kernel:
299	if (fixup_exception(regs))
300		return;
301
302	tsk->thread.error_code = error_code;
303	tsk->thread.trap_no = 13;
304	if (notify_die(DIE_GPF, "general protection fault", regs,
305				error_code, 13, SIGSEGV) == NOTIFY_STOP)
306		return;
307	die("general protection fault", regs, error_code);
308}
309
310static int __init setup_unknown_nmi_panic(char *str)
311{
312	unknown_nmi_panic = 1;
313	return 1;
314}
315__setup("unknown_nmi_panic", setup_unknown_nmi_panic);
316
317static notrace __kprobes void
318pci_serr_error(unsigned char reason, struct pt_regs *regs)
319{
320	pr_emerg("NMI: PCI system error (SERR) for reason %02x on CPU %d.\n",
321		 reason, smp_processor_id());
322
 
323	/*
324	 * On some machines, PCI SERR line is used to report memory
325	 * errors. EDAC makes use of it.
326	 */
327#if defined(CONFIG_EDAC)
328	if (edac_handler_set()) {
329		edac_atomic_assert_error();
330		return;
331	}
332#endif
333
334	if (panic_on_unrecovered_nmi)
335		panic("NMI: Not continuing");
336
337	pr_emerg("Dazed and confused, but trying to continue\n");
338
339	/* Clear and disable the PCI SERR error line. */
340	reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_SERR;
341	outb(reason, NMI_REASON_PORT);
342}
343
344static notrace __kprobes void
345io_check_error(unsigned char reason, struct pt_regs *regs)
346{
347	unsigned long i;
348
349	pr_emerg(
350	"NMI: IOCK error (debug interrupt?) for reason %02x on CPU %d.\n",
351		 reason, smp_processor_id());
352	show_registers(regs);
353
354	if (panic_on_io_nmi)
355		panic("NMI IOCK error: Not continuing");
356
357	/* Re-enable the IOCK line, wait for a few seconds */
358	reason = (reason & NMI_REASON_CLEAR_MASK) | NMI_REASON_CLEAR_IOCHK;
359	outb(reason, NMI_REASON_PORT);
360
361	i = 20000;
362	while (--i) {
363		touch_nmi_watchdog();
364		udelay(100);
365	}
366
367	reason &= ~NMI_REASON_CLEAR_IOCHK;
368	outb(reason, NMI_REASON_PORT);
369}
370
371static notrace __kprobes void
372unknown_nmi_error(unsigned char reason, struct pt_regs *regs)
373{
374	if (notify_die(DIE_NMIUNKNOWN, "nmi", regs, reason, 2, SIGINT) ==
375			NOTIFY_STOP)
376		return;
377#ifdef CONFIG_MCA
378	/*
379	 * Might actually be able to figure out what the guilty party
380	 * is:
381	 */
382	if (MCA_bus) {
383		mca_handle_nmi();
384		return;
385	}
386#endif
387	pr_emerg("Uhhuh. NMI received for unknown reason %02x on CPU %d.\n",
388		 reason, smp_processor_id());
389
390	pr_emerg("Do you have a strange power saving mode enabled?\n");
391	if (unknown_nmi_panic || panic_on_unrecovered_nmi)
392		panic("NMI: Not continuing");
393
394	pr_emerg("Dazed and confused, but trying to continue\n");
395}
 
 
 
 
396
397static notrace __kprobes void default_do_nmi(struct pt_regs *regs)
398{
399	unsigned char reason = 0;
400
401	/*
402	 * CPU-specific NMI must be processed before non-CPU-specific
403	 * NMI, otherwise we may lose it, because the CPU-specific
404	 * NMI can not be detected/processed on other CPUs.
405	 */
406	if (notify_die(DIE_NMI, "nmi", regs, 0, 2, SIGINT) == NOTIFY_STOP)
407		return;
408
409	/* Non-CPU-specific NMI: NMI sources can be processed on any CPU */
410	raw_spin_lock(&nmi_reason_lock);
411	reason = get_nmi_reason();
412
413	if (reason & NMI_REASON_MASK) {
414		if (reason & NMI_REASON_SERR)
415			pci_serr_error(reason, regs);
416		else if (reason & NMI_REASON_IOCHK)
417			io_check_error(reason, regs);
418#ifdef CONFIG_X86_32
419		/*
420		 * Reassert NMI in case it became active
421		 * meanwhile as it's edge-triggered:
422		 */
423		reassert_nmi();
424#endif
425		raw_spin_unlock(&nmi_reason_lock);
426		return;
427	}
428	raw_spin_unlock(&nmi_reason_lock);
429
430	unknown_nmi_error(reason, regs);
431}
432
433dotraplinkage notrace __kprobes void
434do_nmi(struct pt_regs *regs, long error_code)
435{
436	nmi_enter();
437
438	inc_irq_stat(__nmi_count);
439
440	if (!ignore_nmis)
441		default_do_nmi(regs);
442
443	nmi_exit();
444}
445
446void stop_nmi(void)
447{
448	ignore_nmis++;
449}
450
451void restart_nmi(void)
452{
453	ignore_nmis--;
454}
455
456/* May run on IST stack. */
457dotraplinkage void __kprobes do_int3(struct pt_regs *regs, long error_code)
458{
459#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
460	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
461			== NOTIFY_STOP)
462		return;
463#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
464#ifdef CONFIG_KPROBES
465	if (notify_die(DIE_INT3, "int3", regs, error_code, 3, SIGTRAP)
466			== NOTIFY_STOP)
467		return;
468#else
469	if (notify_die(DIE_TRAP, "int3", regs, error_code, 3, SIGTRAP)
470			== NOTIFY_STOP)
471		return;
472#endif
473
474	preempt_conditional_sti(regs);
475	do_trap(3, SIGTRAP, "int3", regs, error_code, NULL);
476	preempt_conditional_cli(regs);
 
 
 
477}
478
479#ifdef CONFIG_X86_64
480/*
481 * Help handler running on IST stack to switch back to user stack
482 * for scheduling or signal handling. The actual stack switch is done in
483 * entry.S
484 */
485asmlinkage __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
486{
487	struct pt_regs *regs = eregs;
488	/* Did already sync */
489	if (eregs == (struct pt_regs *)eregs->sp)
490		;
491	/* Exception from user space */
492	else if (user_mode(eregs))
493		regs = task_pt_regs(current);
494	/*
495	 * Exception from kernel and interrupts are enabled. Move to
496	 * kernel process stack.
497	 */
498	else if (eregs->flags & X86_EFLAGS_IF)
499		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
500	if (eregs != regs)
501		*regs = *eregs;
502	return regs;
503}
504#endif
505
506/*
507 * Our handling of the processor debug registers is non-trivial.
508 * We do not clear them on entry and exit from the kernel. Therefore
509 * it is possible to get a watchpoint trap here from inside the kernel.
510 * However, the code in ./ptrace.c has ensured that the user can
511 * only set watchpoints on userspace addresses. Therefore the in-kernel
512 * watchpoint trap can only occur in code which is reading/writing
513 * from user space. Such code must not hold kernel locks (since it
514 * can equally take a page fault), therefore it is safe to call
515 * force_sig_info even though that claims and releases locks.
516 *
517 * Code in ./signal.c ensures that the debug control register
518 * is restored before we deliver any signal, and therefore that
519 * user code runs with the correct debug control register even though
520 * we clear it here.
521 *
522 * Being careful here means that we don't have to be as careful in a
523 * lot of more complicated places (task switching can be a bit lazy
524 * about restoring all the debug state, and ptrace doesn't have to
525 * find every occurrence of the TF bit that could be saved away even
526 * by user code)
527 *
528 * May run on IST stack.
529 */
530dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
531{
532	struct task_struct *tsk = current;
 
533	int user_icebp = 0;
534	unsigned long dr6;
535	int si_code;
536
 
 
537	get_debugreg(dr6, 6);
538
539	/* Filter out all the reserved bits which are preset to 1 */
540	dr6 &= ~DR6_RESERVED;
541
542	/*
543	 * If dr6 has no reason to give us about the origin of this trap,
544	 * then it's very likely the result of an icebp/int01 trap.
545	 * User wants a sigtrap for that.
546	 */
547	if (!dr6 && user_mode(regs))
548		user_icebp = 1;
549
550	/* Catch kmemcheck conditions first of all! */
551	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
552		return;
553
554	/* DR6 may or may not be cleared by the CPU */
555	set_debugreg(0, 6);
556
557	/*
558	 * The processor cleared BTF, so don't mark that we need it set.
559	 */
560	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
561
562	/* Store the virtualized DR6 value */
563	tsk->thread.debugreg6 = dr6;
564
565	if (notify_die(DIE_DEBUG, "debug", regs, PTR_ERR(&dr6), error_code,
566							SIGTRAP) == NOTIFY_STOP)
567		return;
 
 
 
 
 
 
568
569	/* It's safe to allow irq's after DR6 has been saved */
570	preempt_conditional_sti(regs);
571
572	if (regs->flags & X86_VM_MASK) {
573		handle_vm86_trap((struct kernel_vm86_regs *) regs,
574				error_code, 1);
575		preempt_conditional_cli(regs);
576		return;
 
577	}
578
579	/*
580	 * Single-stepping through system calls: ignore any exceptions in
581	 * kernel space, but re-enable TF when returning to user mode.
582	 *
583	 * We already checked v86 mode above, so we can check for kernel mode
584	 * by just checking the CPL of CS.
585	 */
586	if ((dr6 & DR_STEP) && !user_mode(regs)) {
587		tsk->thread.debugreg6 &= ~DR_STEP;
588		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
589		regs->flags &= ~X86_EFLAGS_TF;
590	}
591	si_code = get_si_code(tsk->thread.debugreg6);
592	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
593		send_sigtrap(tsk, regs, error_code, si_code);
594	preempt_conditional_cli(regs);
 
595
596	return;
 
597}
598
599/*
600 * Note that we play around with the 'TS' bit in an attempt to get
601 * the correct behaviour even in the presence of the asynchronous
602 * IRQ13 behaviour
603 */
604void math_error(struct pt_regs *regs, int error_code, int trapnr)
605{
606	struct task_struct *task = current;
607	siginfo_t info;
608	unsigned short err;
609	char *str = (trapnr == 16) ? "fpu exception" : "simd exception";
 
610
611	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
612		return;
613	conditional_sti(regs);
614
615	if (!user_mode_vm(regs))
616	{
617		if (!fixup_exception(regs)) {
618			task->thread.error_code = error_code;
619			task->thread.trap_no = trapnr;
620			die(str, regs, error_code);
621		}
622		return;
623	}
624
625	/*
626	 * Save the info for the exception handler and clear the error.
627	 */
628	save_init_fpu(task);
629	task->thread.trap_no = trapnr;
630	task->thread.error_code = error_code;
631	info.si_signo = SIGFPE;
632	info.si_errno = 0;
633	info.si_addr = (void __user *)regs->ip;
634	if (trapnr == 16) {
635		unsigned short cwd, swd;
636		/*
637		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
638		 * status.  0x3f is the exception bits in these regs, 0x200 is the
639		 * C1 reg you need in case of a stack fault, 0x040 is the stack
640		 * fault bit.  We should only be taking one exception at a time,
641		 * so if this combination doesn't produce any single exception,
642		 * then we have a bad program that isn't synchronizing its FPU usage
643		 * and it will suffer the consequences since we won't be able to
644		 * fully reproduce the context of the exception
645		 */
646		cwd = get_fpu_cwd(task);
647		swd = get_fpu_swd(task);
648
649		err = swd & ~cwd;
650	} else {
651		/*
652		 * The SIMD FPU exceptions are handled a little differently, as there
653		 * is only a single status/control register.  Thus, to determine which
654		 * unmasked exception was caught we must mask the exception mask bits
655		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
656		 */
657		unsigned short mxcsr = get_fpu_mxcsr(task);
658		err = ~(mxcsr >> 7) & mxcsr;
659	}
660
661	if (err & 0x001) {	/* Invalid op */
662		/*
663		 * swd & 0x240 == 0x040: Stack Underflow
664		 * swd & 0x240 == 0x240: Stack Overflow
665		 * User must clear the SF bit (0x40) if set
666		 */
667		info.si_code = FPE_FLTINV;
668	} else if (err & 0x004) { /* Divide by Zero */
669		info.si_code = FPE_FLTDIV;
670	} else if (err & 0x008) { /* Overflow */
671		info.si_code = FPE_FLTOVF;
672	} else if (err & 0x012) { /* Denormal, Underflow */
673		info.si_code = FPE_FLTUND;
674	} else if (err & 0x020) { /* Precision */
675		info.si_code = FPE_FLTRES;
676	} else {
677		/*
678		 * If we're using IRQ 13, or supposedly even some trap 16
679		 * implementations, it's possible we get a spurious trap...
 
680		 */
681		return;		/* Spurious trap, no error */
682	}
683	force_sig_info(SIGFPE, &info, task);
684}
685
686dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
687{
688#ifdef CONFIG_X86_32
689	ignore_fpu_irq = 1;
690#endif
691
692	math_error(regs, error_code, 16);
 
 
693}
694
695dotraplinkage void
696do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
697{
698	math_error(regs, error_code, 19);
 
 
 
 
699}
700
701dotraplinkage void
702do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
703{
704	conditional_sti(regs);
705#if 0
706	/* No need to warn about this any longer. */
707	printk(KERN_INFO "Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
708#endif
709}
710
711asmlinkage void __attribute__((weak)) smp_thermal_interrupt(void)
712{
713}
714
715asmlinkage void __attribute__((weak)) smp_threshold_interrupt(void)
716{
717}
718
719/*
720 * __math_state_restore assumes that cr0.TS is already clear and the
721 * fpu state is all ready for use.  Used during context switch.
722 */
723void __math_state_restore(void)
724{
725	struct thread_info *thread = current_thread_info();
726	struct task_struct *tsk = thread->task;
727
728	/*
729	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
730	 */
731	if (unlikely(restore_fpu_checking(tsk))) {
732		stts();
733		force_sig(SIGSEGV, tsk);
734		return;
735	}
736
737	thread->status |= TS_USEDFPU;	/* So we fnsave on switch_to() */
738	tsk->fpu_counter++;
739}
740
741/*
742 * 'math_state_restore()' saves the current math information in the
743 * old math state array, and gets the new ones from the current task
744 *
745 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
746 * Don't touch unless you *really* know how it works.
747 *
748 * Must be called with kernel preemption disabled (in this case,
749 * local interrupts are disabled at the call-site in entry.S).
750 */
751asmlinkage void math_state_restore(void)
752{
753	struct thread_info *thread = current_thread_info();
754	struct task_struct *tsk = thread->task;
755
756	if (!tsk_used_math(tsk)) {
757		local_irq_enable();
758		/*
759		 * does a slab alloc which can sleep
760		 */
761		if (init_fpu(tsk)) {
762			/*
763			 * ran out of memory!
764			 */
765			do_group_exit(SIGKILL);
766			return;
767		}
768		local_irq_disable();
769	}
770
771	clts();				/* Allow maths ops (or we recurse) */
772
773	__math_state_restore();
 
 
 
 
 
 
 
 
 
774}
775EXPORT_SYMBOL_GPL(math_state_restore);
776
777dotraplinkage void __kprobes
778do_device_not_available(struct pt_regs *regs, long error_code)
779{
 
 
 
 
 
780#ifdef CONFIG_MATH_EMULATION
781	if (read_cr0() & X86_CR0_EM) {
782		struct math_emu_info info = { };
783
784		conditional_sti(regs);
785
786		info.regs = regs;
787		math_emulate(&info);
 
788		return;
789	}
790#endif
791	math_state_restore(); /* interrupts still off */
792#ifdef CONFIG_X86_32
793	conditional_sti(regs);
794#endif
 
795}
796
797#ifdef CONFIG_X86_32
798dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
799{
800	siginfo_t info;
 
 
 
801	local_irq_enable();
802
803	info.si_signo = SIGILL;
804	info.si_errno = 0;
805	info.si_code = ILL_BADSTK;
806	info.si_addr = NULL;
807	if (notify_die(DIE_TRAP, "iret exception",
808			regs, error_code, 32, SIGILL) == NOTIFY_STOP)
809		return;
810	do_trap(32, SIGILL, "iret exception", regs, error_code, &info);
 
 
811}
812#endif
813
814/* Set of traps needed for early debugging. */
815void __init early_trap_init(void)
816{
817	set_intr_gate_ist(1, &debug, DEBUG_STACK);
818	/* int3 can be called from all */
819	set_system_intr_gate_ist(3, &int3, DEBUG_STACK);
820	set_intr_gate(14, &page_fault);
 
 
821	load_idt(&idt_descr);
822}
823
 
 
 
 
 
 
 
824void __init trap_init(void)
825{
826	int i;
827
828#ifdef CONFIG_EISA
829	void __iomem *p = early_ioremap(0x0FFFD9, 4);
830
831	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
832		EISA_bus = 1;
833	early_iounmap(p, 4);
834#endif
835
836	set_intr_gate(0, &divide_error);
837	set_intr_gate_ist(2, &nmi, NMI_STACK);
838	/* int4 can be called from all */
839	set_system_intr_gate(4, &overflow);
840	set_intr_gate(5, &bounds);
841	set_intr_gate(6, &invalid_op);
842	set_intr_gate(7, &device_not_available);
843#ifdef CONFIG_X86_32
844	set_task_gate(8, GDT_ENTRY_DOUBLEFAULT_TSS);
845#else
846	set_intr_gate_ist(8, &double_fault, DOUBLEFAULT_STACK);
847#endif
848	set_intr_gate(9, &coprocessor_segment_overrun);
849	set_intr_gate(10, &invalid_TSS);
850	set_intr_gate(11, &segment_not_present);
851	set_intr_gate_ist(12, &stack_segment, STACKFAULT_STACK);
852	set_intr_gate(13, &general_protection);
853	set_intr_gate(15, &spurious_interrupt_bug);
854	set_intr_gate(16, &coprocessor_error);
855	set_intr_gate(17, &alignment_check);
856#ifdef CONFIG_X86_MCE
857	set_intr_gate_ist(18, &machine_check, MCE_STACK);
858#endif
859	set_intr_gate(19, &simd_coprocessor_error);
860
861	/* Reserve all the builtin and the syscall vector: */
862	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
863		set_bit(i, used_vectors);
864
865#ifdef CONFIG_IA32_EMULATION
866	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
867	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
868#endif
869
870#ifdef CONFIG_X86_32
871	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
872	set_bit(SYSCALL_VECTOR, used_vectors);
873#endif
874
875	/*
 
 
 
 
 
 
 
 
876	 * Should be a barrier for any external CPU state:
877	 */
878	cpu_init();
879
880	x86_init.irqs.trap_init();
 
 
 
 
 
 
881}
v3.15
  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. */
 69gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
 70#else
 71#include <asm/processor-flags.h>
 72#include <asm/setup.h>
 73
 74asmlinkage int system_call(void);
 
 
 
 
 
 
 
 
 
 75#endif
 76
 77/* Must be page-aligned because the real IDT is used in a fixmap. */
 78gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
 79
 80DECLARE_BITMAP(used_vectors, NR_VECTORS);
 81EXPORT_SYMBOL_GPL(used_vectors);
 82
 
 
 
 
 
 
 
 
 
 83static inline void conditional_sti(struct pt_regs *regs)
 84{
 85	if (regs->flags & X86_EFLAGS_IF)
 86		local_irq_enable();
 87}
 88
 89static 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
 96static inline void conditional_cli(struct pt_regs *regs)
 97{
 98	if (regs->flags & X86_EFLAGS_IF)
 99		local_irq_disable();
100}
101
102static 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
109static int __kprobes
110do_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
139static void __kprobes
140do_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)				\
178dotraplinkage 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)		\
194dotraplinkage 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
214DO_ERROR_INFO(X86_TRAP_DE,     SIGFPE,  "divide error",			divide_error,		     FPE_INTDIV, regs->ip )
215DO_ERROR     (X86_TRAP_OF,     SIGSEGV, "overflow",			overflow					  )
216DO_ERROR     (X86_TRAP_BR,     SIGSEGV, "bounds",			bounds						  )
217DO_ERROR_INFO(X86_TRAP_UD,     SIGILL,  "invalid opcode",		invalid_op,		     ILL_ILLOPN, regs->ip )
218DO_ERROR     (X86_TRAP_OLD_MF, SIGFPE,  "coprocessor segment overrun",	coprocessor_segment_overrun			  )
219DO_ERROR     (X86_TRAP_TS,     SIGSEGV, "invalid TSS",			invalid_TSS					  )
220DO_ERROR     (X86_TRAP_NP,     SIGBUS,  "segment not present",		segment_not_present				  )
221#ifdef CONFIG_X86_32
222DO_ERROR     (X86_TRAP_SS,     SIGBUS,  "stack segment",		stack_segment					  )
223#endif
224DO_ERROR_INFO(X86_TRAP_AC,     SIGBUS,  "alignment check",		alignment_check,	     BUS_ADRALN, 0	  )
225
226#ifdef CONFIG_X86_64
227/* Runs on IST stack */
228dotraplinkage void do_stack_segment(struct pt_regs *regs, long error_code)
229{
230	enum ctx_state prev_state;
231
232	prev_state = exception_enter();
233	if (notify_die(DIE_TRAP, "stack segment", regs, error_code,
234		       X86_TRAP_SS, SIGBUS) != NOTIFY_STOP) {
235		preempt_conditional_sti(regs);
236		do_trap(X86_TRAP_SS, SIGBUS, "stack segment", regs, error_code, NULL);
237		preempt_conditional_cli(regs);
238	}
239	exception_exit(prev_state);
240}
241
242dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
243{
244	static const char str[] = "double fault";
245	struct task_struct *tsk = current;
246
247	exception_enter();
248	/* Return not checked because double check cannot be ignored */
249	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
250
251	tsk->thread.error_code = error_code;
252	tsk->thread.trap_nr = X86_TRAP_DF;
253
254#ifdef CONFIG_DOUBLEFAULT
255	df_debug(regs, error_code);
256#endif
257	/*
258	 * This is always a kernel trap and never fixable (and thus must
259	 * never return).
260	 */
261	for (;;)
262		die(str, regs, error_code);
263}
264#endif
265
266dotraplinkage void __kprobes
267do_general_protection(struct pt_regs *regs, long error_code)
268{
269	struct task_struct *tsk;
270	enum ctx_state prev_state;
271
272	prev_state = exception_enter();
273	conditional_sti(regs);
274
275#ifdef CONFIG_X86_32
276	if (regs->flags & X86_VM_MASK) {
277		local_irq_enable();
278		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
279		goto exit;
280	}
281#endif
282
283	tsk = current;
284	if (!user_mode(regs)) {
285		if (fixup_exception(regs))
286			goto exit;
287
288		tsk->thread.error_code = error_code;
289		tsk->thread.trap_nr = X86_TRAP_GP;
290		if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
291			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
292			die("general protection fault", regs, error_code);
293		goto exit;
294	}
295
296	tsk->thread.error_code = error_code;
297	tsk->thread.trap_nr = X86_TRAP_GP;
298
299	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
300			printk_ratelimit()) {
301		pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
 
302			tsk->comm, task_pid_nr(tsk),
303			regs->ip, regs->sp, error_code);
304		print_vma_addr(" in ", regs->ip);
305		pr_cont("\n");
306	}
307
308	force_sig(SIGSEGV, tsk);
309exit:
310	exception_exit(prev_state);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
311}
 
312
313/* May run on IST stack. */
314dotraplinkage void __kprobes notrace do_int3(struct pt_regs *regs, long error_code)
315{
316	enum ctx_state prev_state;
 
317
318#ifdef CONFIG_DYNAMIC_FTRACE
319	/*
320	 * ftrace must be first, everything else may cause a recursive crash.
321	 * See note by declaration of modifying_ftrace_code in ftrace.c
322	 */
323	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
324	    ftrace_int3_handler(regs))
 
325		return;
 
326#endif
327	if (poke_int3_handler(regs))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
328		return;
 
 
 
 
 
 
 
 
329
330	prev_state = exception_enter();
331#ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
332	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
333				SIGTRAP) == NOTIFY_STOP)
334		goto exit;
335#endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
336
337	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
338			SIGTRAP) == NOTIFY_STOP)
339		goto exit;
340
341	/*
342	 * Let others (NMI) know that the debug stack is in use
343	 * as we may switch to the interrupt stack.
 
344	 */
345	debug_stack_usage_inc();
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
346	preempt_conditional_sti(regs);
347	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
348	preempt_conditional_cli(regs);
349	debug_stack_usage_dec();
350exit:
351	exception_exit(prev_state);
352}
353
354#ifdef CONFIG_X86_64
355/*
356 * Help handler running on IST stack to switch back to user stack
357 * for scheduling or signal handling. The actual stack switch is done in
358 * entry.S
359 */
360asmlinkage __visible __kprobes struct pt_regs *sync_regs(struct pt_regs *eregs)
361{
362	struct pt_regs *regs = eregs;
363	/* Did already sync */
364	if (eregs == (struct pt_regs *)eregs->sp)
365		;
366	/* Exception from user space */
367	else if (user_mode(eregs))
368		regs = task_pt_regs(current);
369	/*
370	 * Exception from kernel and interrupts are enabled. Move to
371	 * kernel process stack.
372	 */
373	else if (eregs->flags & X86_EFLAGS_IF)
374		regs = (struct pt_regs *)(eregs->sp -= sizeof(struct pt_regs));
375	if (eregs != regs)
376		*regs = *eregs;
377	return regs;
378}
379#endif
380
381/*
382 * Our handling of the processor debug registers is non-trivial.
383 * We do not clear them on entry and exit from the kernel. Therefore
384 * it is possible to get a watchpoint trap here from inside the kernel.
385 * However, the code in ./ptrace.c has ensured that the user can
386 * only set watchpoints on userspace addresses. Therefore the in-kernel
387 * watchpoint trap can only occur in code which is reading/writing
388 * from user space. Such code must not hold kernel locks (since it
389 * can equally take a page fault), therefore it is safe to call
390 * force_sig_info even though that claims and releases locks.
391 *
392 * Code in ./signal.c ensures that the debug control register
393 * is restored before we deliver any signal, and therefore that
394 * user code runs with the correct debug control register even though
395 * we clear it here.
396 *
397 * Being careful here means that we don't have to be as careful in a
398 * lot of more complicated places (task switching can be a bit lazy
399 * about restoring all the debug state, and ptrace doesn't have to
400 * find every occurrence of the TF bit that could be saved away even
401 * by user code)
402 *
403 * May run on IST stack.
404 */
405dotraplinkage void __kprobes do_debug(struct pt_regs *regs, long error_code)
406{
407	struct task_struct *tsk = current;
408	enum ctx_state prev_state;
409	int user_icebp = 0;
410	unsigned long dr6;
411	int si_code;
412
413	prev_state = exception_enter();
414
415	get_debugreg(dr6, 6);
416
417	/* Filter out all the reserved bits which are preset to 1 */
418	dr6 &= ~DR6_RESERVED;
419
420	/*
421	 * If dr6 has no reason to give us about the origin of this trap,
422	 * then it's very likely the result of an icebp/int01 trap.
423	 * User wants a sigtrap for that.
424	 */
425	if (!dr6 && user_mode(regs))
426		user_icebp = 1;
427
428	/* Catch kmemcheck conditions first of all! */
429	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
430		goto exit;
431
432	/* DR6 may or may not be cleared by the CPU */
433	set_debugreg(0, 6);
434
435	/*
436	 * The processor cleared BTF, so don't mark that we need it set.
437	 */
438	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
439
440	/* Store the virtualized DR6 value */
441	tsk->thread.debugreg6 = dr6;
442
443	if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
444							SIGTRAP) == NOTIFY_STOP)
445		goto exit;
446
447	/*
448	 * Let others (NMI) know that the debug stack is in use
449	 * as we may switch to the interrupt stack.
450	 */
451	debug_stack_usage_inc();
452
453	/* It's safe to allow irq's after DR6 has been saved */
454	preempt_conditional_sti(regs);
455
456	if (regs->flags & X86_VM_MASK) {
457		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
458					X86_TRAP_DB);
459		preempt_conditional_cli(regs);
460		debug_stack_usage_dec();
461		goto exit;
462	}
463
464	/*
465	 * Single-stepping through system calls: ignore any exceptions in
466	 * kernel space, but re-enable TF when returning to user mode.
467	 *
468	 * We already checked v86 mode above, so we can check for kernel mode
469	 * by just checking the CPL of CS.
470	 */
471	if ((dr6 & DR_STEP) && !user_mode(regs)) {
472		tsk->thread.debugreg6 &= ~DR_STEP;
473		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
474		regs->flags &= ~X86_EFLAGS_TF;
475	}
476	si_code = get_si_code(tsk->thread.debugreg6);
477	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
478		send_sigtrap(tsk, regs, error_code, si_code);
479	preempt_conditional_cli(regs);
480	debug_stack_usage_dec();
481
482exit:
483	exception_exit(prev_state);
484}
485
486/*
487 * Note that we play around with the 'TS' bit in an attempt to get
488 * the correct behaviour even in the presence of the asynchronous
489 * IRQ13 behaviour
490 */
491void math_error(struct pt_regs *regs, int error_code, int trapnr)
492{
493	struct task_struct *task = current;
494	siginfo_t info;
495	unsigned short err;
496	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
497						"simd exception";
498
499	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
500		return;
501	conditional_sti(regs);
502
503	if (!user_mode_vm(regs))
504	{
505		if (!fixup_exception(regs)) {
506			task->thread.error_code = error_code;
507			task->thread.trap_nr = trapnr;
508			die(str, regs, error_code);
509		}
510		return;
511	}
512
513	/*
514	 * Save the info for the exception handler and clear the error.
515	 */
516	save_init_fpu(task);
517	task->thread.trap_nr = trapnr;
518	task->thread.error_code = error_code;
519	info.si_signo = SIGFPE;
520	info.si_errno = 0;
521	info.si_addr = (void __user *)regs->ip;
522	if (trapnr == X86_TRAP_MF) {
523		unsigned short cwd, swd;
524		/*
525		 * (~cwd & swd) will mask out exceptions that are not set to unmasked
526		 * status.  0x3f is the exception bits in these regs, 0x200 is the
527		 * C1 reg you need in case of a stack fault, 0x040 is the stack
528		 * fault bit.  We should only be taking one exception at a time,
529		 * so if this combination doesn't produce any single exception,
530		 * then we have a bad program that isn't synchronizing its FPU usage
531		 * and it will suffer the consequences since we won't be able to
532		 * fully reproduce the context of the exception
533		 */
534		cwd = get_fpu_cwd(task);
535		swd = get_fpu_swd(task);
536
537		err = swd & ~cwd;
538	} else {
539		/*
540		 * The SIMD FPU exceptions are handled a little differently, as there
541		 * is only a single status/control register.  Thus, to determine which
542		 * unmasked exception was caught we must mask the exception mask bits
543		 * at 0x1f80, and then use these to mask the exception bits at 0x3f.
544		 */
545		unsigned short mxcsr = get_fpu_mxcsr(task);
546		err = ~(mxcsr >> 7) & mxcsr;
547	}
548
549	if (err & 0x001) {	/* Invalid op */
550		/*
551		 * swd & 0x240 == 0x040: Stack Underflow
552		 * swd & 0x240 == 0x240: Stack Overflow
553		 * User must clear the SF bit (0x40) if set
554		 */
555		info.si_code = FPE_FLTINV;
556	} else if (err & 0x004) { /* Divide by Zero */
557		info.si_code = FPE_FLTDIV;
558	} else if (err & 0x008) { /* Overflow */
559		info.si_code = FPE_FLTOVF;
560	} else if (err & 0x012) { /* Denormal, Underflow */
561		info.si_code = FPE_FLTUND;
562	} else if (err & 0x020) { /* Precision */
563		info.si_code = FPE_FLTRES;
564	} else {
565		/*
566		 * If we're using IRQ 13, or supposedly even some trap
567		 * X86_TRAP_MF implementations, it's possible
568		 * we get a spurious trap, which is not an error.
569		 */
570		return;
571	}
572	force_sig_info(SIGFPE, &info, task);
573}
574
575dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
576{
577	enum ctx_state prev_state;
 
 
578
579	prev_state = exception_enter();
580	math_error(regs, error_code, X86_TRAP_MF);
581	exception_exit(prev_state);
582}
583
584dotraplinkage void
585do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
586{
587	enum ctx_state prev_state;
588
589	prev_state = exception_enter();
590	math_error(regs, error_code, X86_TRAP_XF);
591	exception_exit(prev_state);
592}
593
594dotraplinkage void
595do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
596{
597	conditional_sti(regs);
598#if 0
599	/* No need to warn about this any longer. */
600	pr_info("Ignoring P6 Local APIC Spurious Interrupt Bug...\n");
601#endif
602}
603
604asmlinkage __visible void __attribute__((weak)) smp_thermal_interrupt(void)
605{
606}
607
608asmlinkage __visible void __attribute__((weak)) smp_threshold_interrupt(void)
609{
610}
611
612/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
613 * 'math_state_restore()' saves the current math information in the
614 * old math state array, and gets the new ones from the current task
615 *
616 * Careful.. There are problems with IBM-designed IRQ13 behaviour.
617 * Don't touch unless you *really* know how it works.
618 *
619 * Must be called with kernel preemption disabled (eg with local
620 * local interrupts as in the case of do_device_not_available).
621 */
622void math_state_restore(void)
623{
624	struct task_struct *tsk = current;
 
625
626	if (!tsk_used_math(tsk)) {
627		local_irq_enable();
628		/*
629		 * does a slab alloc which can sleep
630		 */
631		if (init_fpu(tsk)) {
632			/*
633			 * ran out of memory!
634			 */
635			do_group_exit(SIGKILL);
636			return;
637		}
638		local_irq_disable();
639	}
640
641	__thread_fpu_begin(tsk);
642
643	/*
644	 * Paranoid restore. send a SIGSEGV if we fail to restore the state.
645	 */
646	if (unlikely(restore_fpu_checking(tsk))) {
647		drop_init_fpu(tsk);
648		force_sig(SIGSEGV, tsk);
649		return;
650	}
651
652	tsk->thread.fpu_counter++;
653}
654EXPORT_SYMBOL_GPL(math_state_restore);
655
656dotraplinkage void __kprobes
657do_device_not_available(struct pt_regs *regs, long error_code)
658{
659	enum ctx_state prev_state;
660
661	prev_state = exception_enter();
662	BUG_ON(use_eager_fpu());
663
664#ifdef CONFIG_MATH_EMULATION
665	if (read_cr0() & X86_CR0_EM) {
666		struct math_emu_info info = { };
667
668		conditional_sti(regs);
669
670		info.regs = regs;
671		math_emulate(&info);
672		exception_exit(prev_state);
673		return;
674	}
675#endif
676	math_state_restore(); /* interrupts still off */
677#ifdef CONFIG_X86_32
678	conditional_sti(regs);
679#endif
680	exception_exit(prev_state);
681}
682
683#ifdef CONFIG_X86_32
684dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
685{
686	siginfo_t info;
687	enum ctx_state prev_state;
688
689	prev_state = exception_enter();
690	local_irq_enable();
691
692	info.si_signo = SIGILL;
693	info.si_errno = 0;
694	info.si_code = ILL_BADSTK;
695	info.si_addr = NULL;
696	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
697			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
698		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
699			&info);
700	}
701	exception_exit(prev_state);
702}
703#endif
704
705/* Set of traps needed for early debugging. */
706void __init early_trap_init(void)
707{
708	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
709	/* int3 can be called from all */
710	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
711#ifdef CONFIG_X86_32
712	set_intr_gate(X86_TRAP_PF, page_fault);
713#endif
714	load_idt(&idt_descr);
715}
716
717void __init early_trap_pf_init(void)
718{
719#ifdef CONFIG_X86_64
720	set_intr_gate(X86_TRAP_PF, page_fault);
721#endif
722}
723
724void __init trap_init(void)
725{
726	int i;
727
728#ifdef CONFIG_EISA
729	void __iomem *p = early_ioremap(0x0FFFD9, 4);
730
731	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
732		EISA_bus = 1;
733	early_iounmap(p, 4);
734#endif
735
736	set_intr_gate(X86_TRAP_DE, divide_error);
737	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
738	/* int4 can be called from all */
739	set_system_intr_gate(X86_TRAP_OF, &overflow);
740	set_intr_gate(X86_TRAP_BR, bounds);
741	set_intr_gate(X86_TRAP_UD, invalid_op);
742	set_intr_gate(X86_TRAP_NM, device_not_available);
743#ifdef CONFIG_X86_32
744	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
745#else
746	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
747#endif
748	set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
749	set_intr_gate(X86_TRAP_TS, invalid_TSS);
750	set_intr_gate(X86_TRAP_NP, segment_not_present);
751	set_intr_gate_ist(X86_TRAP_SS, &stack_segment, STACKFAULT_STACK);
752	set_intr_gate(X86_TRAP_GP, general_protection);
753	set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
754	set_intr_gate(X86_TRAP_MF, coprocessor_error);
755	set_intr_gate(X86_TRAP_AC, alignment_check);
756#ifdef CONFIG_X86_MCE
757	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
758#endif
759	set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
760
761	/* Reserve all the builtin and the syscall vector: */
762	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
763		set_bit(i, used_vectors);
764
765#ifdef CONFIG_IA32_EMULATION
766	set_system_intr_gate(IA32_SYSCALL_VECTOR, ia32_syscall);
767	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
768#endif
769
770#ifdef CONFIG_X86_32
771	set_system_trap_gate(SYSCALL_VECTOR, &system_call);
772	set_bit(SYSCALL_VECTOR, used_vectors);
773#endif
774
775	/*
776	 * Set the IDT descriptor to a fixed read-only location, so that the
777	 * "sidt" instruction will not leak the location of the kernel, and
778	 * to defend the IDT against arbitrary memory write vulnerabilities.
779	 * It will be reloaded in cpu_init() */
780	__set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
781	idt_descr.address = fix_to_virt(FIX_RO_IDT);
782
783	/*
784	 * Should be a barrier for any external CPU state:
785	 */
786	cpu_init();
787
788	x86_init.irqs.trap_init();
789
790#ifdef CONFIG_X86_64
791	memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
792	set_nmi_gate(X86_TRAP_DB, &debug);
793	set_nmi_gate(X86_TRAP_BP, &int3);
794#endif
795}