xref: /linux/arch/x86/kernel/traps.c (revision e5c86679d5e864947a52fb31e45a425dea3e7fa9)
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/export.h>
25 #include <linux/ptrace.h>
26 #include <linux/uprobes.h>
27 #include <linux/string.h>
28 #include <linux/delay.h>
29 #include <linux/errno.h>
30 #include <linux/kexec.h>
31 #include <linux/sched.h>
32 #include <linux/sched/task_stack.h>
33 #include <linux/timer.h>
34 #include <linux/init.h>
35 #include <linux/bug.h>
36 #include <linux/nmi.h>
37 #include <linux/mm.h>
38 #include <linux/smp.h>
39 #include <linux/io.h>
40 
41 #ifdef CONFIG_EISA
42 #include <linux/ioport.h>
43 #include <linux/eisa.h>
44 #endif
45 
46 #if defined(CONFIG_EDAC)
47 #include <linux/edac.h>
48 #endif
49 
50 #include <asm/kmemcheck.h>
51 #include <asm/stacktrace.h>
52 #include <asm/processor.h>
53 #include <asm/debugreg.h>
54 #include <linux/atomic.h>
55 #include <asm/text-patching.h>
56 #include <asm/ftrace.h>
57 #include <asm/traps.h>
58 #include <asm/desc.h>
59 #include <asm/fpu/internal.h>
60 #include <asm/mce.h>
61 #include <asm/fixmap.h>
62 #include <asm/mach_traps.h>
63 #include <asm/alternative.h>
64 #include <asm/fpu/xstate.h>
65 #include <asm/trace/mpx.h>
66 #include <asm/mpx.h>
67 #include <asm/vm86.h>
68 
69 #ifdef CONFIG_X86_64
70 #include <asm/x86_init.h>
71 #include <asm/pgalloc.h>
72 #include <asm/proto.h>
73 
74 /* No need to be aligned, but done to keep all IDTs defined the same way. */
75 gate_desc debug_idt_table[NR_VECTORS] __page_aligned_bss;
76 #else
77 #include <asm/processor-flags.h>
78 #include <asm/setup.h>
79 #include <asm/proto.h>
80 #endif
81 
82 /* Must be page-aligned because the real IDT is used in a fixmap. */
83 gate_desc idt_table[NR_VECTORS] __page_aligned_bss;
84 
85 DECLARE_BITMAP(used_vectors, NR_VECTORS);
86 EXPORT_SYMBOL_GPL(used_vectors);
87 
88 static inline void cond_local_irq_enable(struct pt_regs *regs)
89 {
90 	if (regs->flags & X86_EFLAGS_IF)
91 		local_irq_enable();
92 }
93 
94 static inline void cond_local_irq_disable(struct pt_regs *regs)
95 {
96 	if (regs->flags & X86_EFLAGS_IF)
97 		local_irq_disable();
98 }
99 
100 /*
101  * In IST context, we explicitly disable preemption.  This serves two
102  * purposes: it makes it much less likely that we would accidentally
103  * schedule in IST context and it will force a warning if we somehow
104  * manage to schedule by accident.
105  */
106 void ist_enter(struct pt_regs *regs)
107 {
108 	if (user_mode(regs)) {
109 		RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
110 	} else {
111 		/*
112 		 * We might have interrupted pretty much anything.  In
113 		 * fact, if we're a machine check, we can even interrupt
114 		 * NMI processing.  We don't want in_nmi() to return true,
115 		 * but we need to notify RCU.
116 		 */
117 		rcu_nmi_enter();
118 	}
119 
120 	preempt_disable();
121 
122 	/* This code is a bit fragile.  Test it. */
123 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
124 }
125 
126 void ist_exit(struct pt_regs *regs)
127 {
128 	preempt_enable_no_resched();
129 
130 	if (!user_mode(regs))
131 		rcu_nmi_exit();
132 }
133 
134 /**
135  * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
136  * @regs:	regs passed to the IST exception handler
137  *
138  * IST exception handlers normally cannot schedule.  As a special
139  * exception, if the exception interrupted userspace code (i.e.
140  * user_mode(regs) would return true) and the exception was not
141  * a double fault, it can be safe to schedule.  ist_begin_non_atomic()
142  * begins a non-atomic section within an ist_enter()/ist_exit() region.
143  * Callers are responsible for enabling interrupts themselves inside
144  * the non-atomic section, and callers must call ist_end_non_atomic()
145  * before ist_exit().
146  */
147 void ist_begin_non_atomic(struct pt_regs *regs)
148 {
149 	BUG_ON(!user_mode(regs));
150 
151 	/*
152 	 * Sanity check: we need to be on the normal thread stack.  This
153 	 * will catch asm bugs and any attempt to use ist_preempt_enable
154 	 * from double_fault.
155 	 */
156 	BUG_ON((unsigned long)(current_top_of_stack() -
157 			       current_stack_pointer()) >= THREAD_SIZE);
158 
159 	preempt_enable_no_resched();
160 }
161 
162 /**
163  * ist_end_non_atomic() - begin a non-atomic section in an IST exception
164  *
165  * Ends a non-atomic section started with ist_begin_non_atomic().
166  */
167 void ist_end_non_atomic(void)
168 {
169 	preempt_disable();
170 }
171 
172 static nokprobe_inline int
173 do_trap_no_signal(struct task_struct *tsk, int trapnr, char *str,
174 		  struct pt_regs *regs,	long error_code)
175 {
176 	if (v8086_mode(regs)) {
177 		/*
178 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
179 		 * On nmi (interrupt 2), do_trap should not be called.
180 		 */
181 		if (trapnr < X86_TRAP_UD) {
182 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
183 						error_code, trapnr))
184 				return 0;
185 		}
186 		return -1;
187 	}
188 
189 	if (!user_mode(regs)) {
190 		if (!fixup_exception(regs, trapnr)) {
191 			tsk->thread.error_code = error_code;
192 			tsk->thread.trap_nr = trapnr;
193 			die(str, regs, error_code);
194 		}
195 		return 0;
196 	}
197 
198 	return -1;
199 }
200 
201 static siginfo_t *fill_trap_info(struct pt_regs *regs, int signr, int trapnr,
202 				siginfo_t *info)
203 {
204 	unsigned long siaddr;
205 	int sicode;
206 
207 	switch (trapnr) {
208 	default:
209 		return SEND_SIG_PRIV;
210 
211 	case X86_TRAP_DE:
212 		sicode = FPE_INTDIV;
213 		siaddr = uprobe_get_trap_addr(regs);
214 		break;
215 	case X86_TRAP_UD:
216 		sicode = ILL_ILLOPN;
217 		siaddr = uprobe_get_trap_addr(regs);
218 		break;
219 	case X86_TRAP_AC:
220 		sicode = BUS_ADRALN;
221 		siaddr = 0;
222 		break;
223 	}
224 
225 	info->si_signo = signr;
226 	info->si_errno = 0;
227 	info->si_code = sicode;
228 	info->si_addr = (void __user *)siaddr;
229 	return info;
230 }
231 
232 static void
233 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
234 	long error_code, siginfo_t *info)
235 {
236 	struct task_struct *tsk = current;
237 
238 
239 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
240 		return;
241 	/*
242 	 * We want error_code and trap_nr set for userspace faults and
243 	 * kernelspace faults which result in die(), but not
244 	 * kernelspace faults which are fixed up.  die() gives the
245 	 * process no chance to handle the signal and notice the
246 	 * kernel fault information, so that won't result in polluting
247 	 * the information about previously queued, but not yet
248 	 * delivered, faults.  See also do_general_protection below.
249 	 */
250 	tsk->thread.error_code = error_code;
251 	tsk->thread.trap_nr = trapnr;
252 
253 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
254 	    printk_ratelimit()) {
255 		pr_info("%s[%d] trap %s ip:%lx sp:%lx error:%lx",
256 			tsk->comm, tsk->pid, str,
257 			regs->ip, regs->sp, error_code);
258 		print_vma_addr(KERN_CONT " in ", regs->ip);
259 		pr_cont("\n");
260 	}
261 
262 	force_sig_info(signr, info ?: SEND_SIG_PRIV, tsk);
263 }
264 NOKPROBE_SYMBOL(do_trap);
265 
266 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
267 			  unsigned long trapnr, int signr)
268 {
269 	siginfo_t info;
270 
271 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
272 
273 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
274 			NOTIFY_STOP) {
275 		cond_local_irq_enable(regs);
276 		do_trap(trapnr, signr, str, regs, error_code,
277 			fill_trap_info(regs, signr, trapnr, &info));
278 	}
279 }
280 
281 #define DO_ERROR(trapnr, signr, str, name)				\
282 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	\
283 {									\
284 	do_error_trap(regs, error_code, str, trapnr, signr);		\
285 }
286 
287 DO_ERROR(X86_TRAP_DE,     SIGFPE,  "divide error",		divide_error)
288 DO_ERROR(X86_TRAP_OF,     SIGSEGV, "overflow",			overflow)
289 DO_ERROR(X86_TRAP_UD,     SIGILL,  "invalid opcode",		invalid_op)
290 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE,  "coprocessor segment overrun",coprocessor_segment_overrun)
291 DO_ERROR(X86_TRAP_TS,     SIGSEGV, "invalid TSS",		invalid_TSS)
292 DO_ERROR(X86_TRAP_NP,     SIGBUS,  "segment not present",	segment_not_present)
293 DO_ERROR(X86_TRAP_SS,     SIGBUS,  "stack segment",		stack_segment)
294 DO_ERROR(X86_TRAP_AC,     SIGBUS,  "alignment check",		alignment_check)
295 
296 #ifdef CONFIG_VMAP_STACK
297 __visible void __noreturn handle_stack_overflow(const char *message,
298 						struct pt_regs *regs,
299 						unsigned long fault_address)
300 {
301 	printk(KERN_EMERG "BUG: stack guard page was hit at %p (stack is %p..%p)\n",
302 		 (void *)fault_address, current->stack,
303 		 (char *)current->stack + THREAD_SIZE - 1);
304 	die(message, regs, 0);
305 
306 	/* Be absolutely certain we don't return. */
307 	panic(message);
308 }
309 #endif
310 
311 #ifdef CONFIG_X86_64
312 /* Runs on IST stack */
313 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
314 {
315 	static const char str[] = "double fault";
316 	struct task_struct *tsk = current;
317 #ifdef CONFIG_VMAP_STACK
318 	unsigned long cr2;
319 #endif
320 
321 #ifdef CONFIG_X86_ESPFIX64
322 	extern unsigned char native_irq_return_iret[];
323 
324 	/*
325 	 * If IRET takes a non-IST fault on the espfix64 stack, then we
326 	 * end up promoting it to a doublefault.  In that case, modify
327 	 * the stack to make it look like we just entered the #GP
328 	 * handler from user space, similar to bad_iret.
329 	 *
330 	 * No need for ist_enter here because we don't use RCU.
331 	 */
332 	if (((long)regs->sp >> PGDIR_SHIFT) == ESPFIX_PGD_ENTRY &&
333 		regs->cs == __KERNEL_CS &&
334 		regs->ip == (unsigned long)native_irq_return_iret)
335 	{
336 		struct pt_regs *normal_regs = task_pt_regs(current);
337 
338 		/* Fake a #GP(0) from userspace. */
339 		memmove(&normal_regs->ip, (void *)regs->sp, 5*8);
340 		normal_regs->orig_ax = 0;  /* Missing (lost) #GP error code */
341 		regs->ip = (unsigned long)general_protection;
342 		regs->sp = (unsigned long)&normal_regs->orig_ax;
343 
344 		return;
345 	}
346 #endif
347 
348 	ist_enter(regs);
349 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
350 
351 	tsk->thread.error_code = error_code;
352 	tsk->thread.trap_nr = X86_TRAP_DF;
353 
354 #ifdef CONFIG_VMAP_STACK
355 	/*
356 	 * If we overflow the stack into a guard page, the CPU will fail
357 	 * to deliver #PF and will send #DF instead.  Similarly, if we
358 	 * take any non-IST exception while too close to the bottom of
359 	 * the stack, the processor will get a page fault while
360 	 * delivering the exception and will generate a double fault.
361 	 *
362 	 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
363 	 * Page-Fault Exception (#PF):
364 	 *
365 	 *   Processors update CR2 whenever a page fault is detected. If a
366 	 *   second page fault occurs while an earlier page fault is being
367 	 *   deliv- ered, the faulting linear address of the second fault will
368 	 *   overwrite the contents of CR2 (replacing the previous
369 	 *   address). These updates to CR2 occur even if the page fault
370 	 *   results in a double fault or occurs during the delivery of a
371 	 *   double fault.
372 	 *
373 	 * The logic below has a small possibility of incorrectly diagnosing
374 	 * some errors as stack overflows.  For example, if the IDT or GDT
375 	 * gets corrupted such that #GP delivery fails due to a bad descriptor
376 	 * causing #GP and we hit this condition while CR2 coincidentally
377 	 * points to the stack guard page, we'll think we overflowed the
378 	 * stack.  Given that we're going to panic one way or another
379 	 * if this happens, this isn't necessarily worth fixing.
380 	 *
381 	 * If necessary, we could improve the test by only diagnosing
382 	 * a stack overflow if the saved RSP points within 47 bytes of
383 	 * the bottom of the stack: if RSP == tsk_stack + 48 and we
384 	 * take an exception, the stack is already aligned and there
385 	 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
386 	 * possible error code, so a stack overflow would *not* double
387 	 * fault.  With any less space left, exception delivery could
388 	 * fail, and, as a practical matter, we've overflowed the
389 	 * stack even if the actual trigger for the double fault was
390 	 * something else.
391 	 */
392 	cr2 = read_cr2();
393 	if ((unsigned long)task_stack_page(tsk) - 1 - cr2 < PAGE_SIZE)
394 		handle_stack_overflow("kernel stack overflow (double-fault)", regs, cr2);
395 #endif
396 
397 #ifdef CONFIG_DOUBLEFAULT
398 	df_debug(regs, error_code);
399 #endif
400 	/*
401 	 * This is always a kernel trap and never fixable (and thus must
402 	 * never return).
403 	 */
404 	for (;;)
405 		die(str, regs, error_code);
406 }
407 #endif
408 
409 dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
410 {
411 	const struct mpx_bndcsr *bndcsr;
412 	siginfo_t *info;
413 
414 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
415 	if (notify_die(DIE_TRAP, "bounds", regs, error_code,
416 			X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
417 		return;
418 	cond_local_irq_enable(regs);
419 
420 	if (!user_mode(regs))
421 		die("bounds", regs, error_code);
422 
423 	if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
424 		/* The exception is not from Intel MPX */
425 		goto exit_trap;
426 	}
427 
428 	/*
429 	 * We need to look at BNDSTATUS to resolve this exception.
430 	 * A NULL here might mean that it is in its 'init state',
431 	 * which is all zeros which indicates MPX was not
432 	 * responsible for the exception.
433 	 */
434 	bndcsr = get_xsave_field_ptr(XFEATURE_MASK_BNDCSR);
435 	if (!bndcsr)
436 		goto exit_trap;
437 
438 	trace_bounds_exception_mpx(bndcsr);
439 	/*
440 	 * The error code field of the BNDSTATUS register communicates status
441 	 * information of a bound range exception #BR or operation involving
442 	 * bound directory.
443 	 */
444 	switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
445 	case 2:	/* Bound directory has invalid entry. */
446 		if (mpx_handle_bd_fault())
447 			goto exit_trap;
448 		break; /* Success, it was handled */
449 	case 1: /* Bound violation. */
450 		info = mpx_generate_siginfo(regs);
451 		if (IS_ERR(info)) {
452 			/*
453 			 * We failed to decode the MPX instruction.  Act as if
454 			 * the exception was not caused by MPX.
455 			 */
456 			goto exit_trap;
457 		}
458 		/*
459 		 * Success, we decoded the instruction and retrieved
460 		 * an 'info' containing the address being accessed
461 		 * which caused the exception.  This information
462 		 * allows and application to possibly handle the
463 		 * #BR exception itself.
464 		 */
465 		do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, info);
466 		kfree(info);
467 		break;
468 	case 0: /* No exception caused by Intel MPX operations. */
469 		goto exit_trap;
470 	default:
471 		die("bounds", regs, error_code);
472 	}
473 
474 	return;
475 
476 exit_trap:
477 	/*
478 	 * This path out is for all the cases where we could not
479 	 * handle the exception in some way (like allocating a
480 	 * table or telling userspace about it.  We will also end
481 	 * up here if the kernel has MPX turned off at compile
482 	 * time..
483 	 */
484 	do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, NULL);
485 }
486 
487 dotraplinkage void
488 do_general_protection(struct pt_regs *regs, long error_code)
489 {
490 	struct task_struct *tsk;
491 
492 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
493 	cond_local_irq_enable(regs);
494 
495 	if (v8086_mode(regs)) {
496 		local_irq_enable();
497 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
498 		return;
499 	}
500 
501 	tsk = current;
502 	if (!user_mode(regs)) {
503 		if (fixup_exception(regs, X86_TRAP_GP))
504 			return;
505 
506 		tsk->thread.error_code = error_code;
507 		tsk->thread.trap_nr = X86_TRAP_GP;
508 		if (notify_die(DIE_GPF, "general protection fault", regs, error_code,
509 			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
510 			die("general protection fault", regs, error_code);
511 		return;
512 	}
513 
514 	tsk->thread.error_code = error_code;
515 	tsk->thread.trap_nr = X86_TRAP_GP;
516 
517 	if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) &&
518 			printk_ratelimit()) {
519 		pr_info("%s[%d] general protection ip:%lx sp:%lx error:%lx",
520 			tsk->comm, task_pid_nr(tsk),
521 			regs->ip, regs->sp, error_code);
522 		print_vma_addr(KERN_CONT " in ", regs->ip);
523 		pr_cont("\n");
524 	}
525 
526 	force_sig_info(SIGSEGV, SEND_SIG_PRIV, tsk);
527 }
528 NOKPROBE_SYMBOL(do_general_protection);
529 
530 /* May run on IST stack. */
531 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
532 {
533 #ifdef CONFIG_DYNAMIC_FTRACE
534 	/*
535 	 * ftrace must be first, everything else may cause a recursive crash.
536 	 * See note by declaration of modifying_ftrace_code in ftrace.c
537 	 */
538 	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
539 	    ftrace_int3_handler(regs))
540 		return;
541 #endif
542 	if (poke_int3_handler(regs))
543 		return;
544 
545 	ist_enter(regs);
546 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
547 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
548 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
549 				SIGTRAP) == NOTIFY_STOP)
550 		goto exit;
551 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
552 
553 #ifdef CONFIG_KPROBES
554 	if (kprobe_int3_handler(regs))
555 		goto exit;
556 #endif
557 
558 	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
559 			SIGTRAP) == NOTIFY_STOP)
560 		goto exit;
561 
562 	/*
563 	 * Let others (NMI) know that the debug stack is in use
564 	 * as we may switch to the interrupt stack.
565 	 */
566 	debug_stack_usage_inc();
567 	cond_local_irq_enable(regs);
568 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, NULL);
569 	cond_local_irq_disable(regs);
570 	debug_stack_usage_dec();
571 exit:
572 	ist_exit(regs);
573 }
574 NOKPROBE_SYMBOL(do_int3);
575 
576 #ifdef CONFIG_X86_64
577 /*
578  * Help handler running on IST stack to switch off the IST stack if the
579  * interrupted code was in user mode. The actual stack switch is done in
580  * entry_64.S
581  */
582 asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
583 {
584 	struct pt_regs *regs = task_pt_regs(current);
585 	*regs = *eregs;
586 	return regs;
587 }
588 NOKPROBE_SYMBOL(sync_regs);
589 
590 struct bad_iret_stack {
591 	void *error_entry_ret;
592 	struct pt_regs regs;
593 };
594 
595 asmlinkage __visible notrace
596 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
597 {
598 	/*
599 	 * This is called from entry_64.S early in handling a fault
600 	 * caused by a bad iret to user mode.  To handle the fault
601 	 * correctly, we want move our stack frame to task_pt_regs
602 	 * and we want to pretend that the exception came from the
603 	 * iret target.
604 	 */
605 	struct bad_iret_stack *new_stack =
606 		container_of(task_pt_regs(current),
607 			     struct bad_iret_stack, regs);
608 
609 	/* Copy the IRET target to the new stack. */
610 	memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
611 
612 	/* Copy the remainder of the stack from the current stack. */
613 	memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
614 
615 	BUG_ON(!user_mode(&new_stack->regs));
616 	return new_stack;
617 }
618 NOKPROBE_SYMBOL(fixup_bad_iret);
619 #endif
620 
621 static bool is_sysenter_singlestep(struct pt_regs *regs)
622 {
623 	/*
624 	 * We don't try for precision here.  If we're anywhere in the region of
625 	 * code that can be single-stepped in the SYSENTER entry path, then
626 	 * assume that this is a useless single-step trap due to SYSENTER
627 	 * being invoked with TF set.  (We don't know in advance exactly
628 	 * which instructions will be hit because BTF could plausibly
629 	 * be set.)
630 	 */
631 #ifdef CONFIG_X86_32
632 	return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
633 		(unsigned long)__end_SYSENTER_singlestep_region -
634 		(unsigned long)__begin_SYSENTER_singlestep_region;
635 #elif defined(CONFIG_IA32_EMULATION)
636 	return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
637 		(unsigned long)__end_entry_SYSENTER_compat -
638 		(unsigned long)entry_SYSENTER_compat;
639 #else
640 	return false;
641 #endif
642 }
643 
644 /*
645  * Our handling of the processor debug registers is non-trivial.
646  * We do not clear them on entry and exit from the kernel. Therefore
647  * it is possible to get a watchpoint trap here from inside the kernel.
648  * However, the code in ./ptrace.c has ensured that the user can
649  * only set watchpoints on userspace addresses. Therefore the in-kernel
650  * watchpoint trap can only occur in code which is reading/writing
651  * from user space. Such code must not hold kernel locks (since it
652  * can equally take a page fault), therefore it is safe to call
653  * force_sig_info even though that claims and releases locks.
654  *
655  * Code in ./signal.c ensures that the debug control register
656  * is restored before we deliver any signal, and therefore that
657  * user code runs with the correct debug control register even though
658  * we clear it here.
659  *
660  * Being careful here means that we don't have to be as careful in a
661  * lot of more complicated places (task switching can be a bit lazy
662  * about restoring all the debug state, and ptrace doesn't have to
663  * find every occurrence of the TF bit that could be saved away even
664  * by user code)
665  *
666  * May run on IST stack.
667  */
668 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
669 {
670 	struct task_struct *tsk = current;
671 	int user_icebp = 0;
672 	unsigned long dr6;
673 	int si_code;
674 
675 	ist_enter(regs);
676 
677 	get_debugreg(dr6, 6);
678 	/*
679 	 * The Intel SDM says:
680 	 *
681 	 *   Certain debug exceptions may clear bits 0-3. The remaining
682 	 *   contents of the DR6 register are never cleared by the
683 	 *   processor. To avoid confusion in identifying debug
684 	 *   exceptions, debug handlers should clear the register before
685 	 *   returning to the interrupted task.
686 	 *
687 	 * Keep it simple: clear DR6 immediately.
688 	 */
689 	set_debugreg(0, 6);
690 
691 	/* Filter out all the reserved bits which are preset to 1 */
692 	dr6 &= ~DR6_RESERVED;
693 
694 	/*
695 	 * The SDM says "The processor clears the BTF flag when it
696 	 * generates a debug exception."  Clear TIF_BLOCKSTEP to keep
697 	 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
698 	 */
699 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
700 
701 	if (unlikely(!user_mode(regs) && (dr6 & DR_STEP) &&
702 		     is_sysenter_singlestep(regs))) {
703 		dr6 &= ~DR_STEP;
704 		if (!dr6)
705 			goto exit;
706 		/*
707 		 * else we might have gotten a single-step trap and hit a
708 		 * watchpoint at the same time, in which case we should fall
709 		 * through and handle the watchpoint.
710 		 */
711 	}
712 
713 	/*
714 	 * If dr6 has no reason to give us about the origin of this trap,
715 	 * then it's very likely the result of an icebp/int01 trap.
716 	 * User wants a sigtrap for that.
717 	 */
718 	if (!dr6 && user_mode(regs))
719 		user_icebp = 1;
720 
721 	/* Catch kmemcheck conditions! */
722 	if ((dr6 & DR_STEP) && kmemcheck_trap(regs))
723 		goto exit;
724 
725 	/* Store the virtualized DR6 value */
726 	tsk->thread.debugreg6 = dr6;
727 
728 #ifdef CONFIG_KPROBES
729 	if (kprobe_debug_handler(regs))
730 		goto exit;
731 #endif
732 
733 	if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
734 							SIGTRAP) == NOTIFY_STOP)
735 		goto exit;
736 
737 	/*
738 	 * Let others (NMI) know that the debug stack is in use
739 	 * as we may switch to the interrupt stack.
740 	 */
741 	debug_stack_usage_inc();
742 
743 	/* It's safe to allow irq's after DR6 has been saved */
744 	cond_local_irq_enable(regs);
745 
746 	if (v8086_mode(regs)) {
747 		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
748 					X86_TRAP_DB);
749 		cond_local_irq_disable(regs);
750 		debug_stack_usage_dec();
751 		goto exit;
752 	}
753 
754 	if (WARN_ON_ONCE((dr6 & DR_STEP) && !user_mode(regs))) {
755 		/*
756 		 * Historical junk that used to handle SYSENTER single-stepping.
757 		 * This should be unreachable now.  If we survive for a while
758 		 * without anyone hitting this warning, we'll turn this into
759 		 * an oops.
760 		 */
761 		tsk->thread.debugreg6 &= ~DR_STEP;
762 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
763 		regs->flags &= ~X86_EFLAGS_TF;
764 	}
765 	si_code = get_si_code(tsk->thread.debugreg6);
766 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
767 		send_sigtrap(tsk, regs, error_code, si_code);
768 	cond_local_irq_disable(regs);
769 	debug_stack_usage_dec();
770 
771 exit:
772 #if defined(CONFIG_X86_32)
773 	/*
774 	 * This is the most likely code path that involves non-trivial use
775 	 * of the SYSENTER stack.  Check that we haven't overrun it.
776 	 */
777 	WARN(this_cpu_read(cpu_tss.SYSENTER_stack_canary) != STACK_END_MAGIC,
778 	     "Overran or corrupted SYSENTER stack\n");
779 #endif
780 	ist_exit(regs);
781 }
782 NOKPROBE_SYMBOL(do_debug);
783 
784 /*
785  * Note that we play around with the 'TS' bit in an attempt to get
786  * the correct behaviour even in the presence of the asynchronous
787  * IRQ13 behaviour
788  */
789 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
790 {
791 	struct task_struct *task = current;
792 	struct fpu *fpu = &task->thread.fpu;
793 	siginfo_t info;
794 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
795 						"simd exception";
796 
797 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, SIGFPE) == NOTIFY_STOP)
798 		return;
799 	cond_local_irq_enable(regs);
800 
801 	if (!user_mode(regs)) {
802 		if (!fixup_exception(regs, trapnr)) {
803 			task->thread.error_code = error_code;
804 			task->thread.trap_nr = trapnr;
805 			die(str, regs, error_code);
806 		}
807 		return;
808 	}
809 
810 	/*
811 	 * Save the info for the exception handler and clear the error.
812 	 */
813 	fpu__save(fpu);
814 
815 	task->thread.trap_nr	= trapnr;
816 	task->thread.error_code = error_code;
817 	info.si_signo		= SIGFPE;
818 	info.si_errno		= 0;
819 	info.si_addr		= (void __user *)uprobe_get_trap_addr(regs);
820 
821 	info.si_code = fpu__exception_code(fpu, trapnr);
822 
823 	/* Retry when we get spurious exceptions: */
824 	if (!info.si_code)
825 		return;
826 
827 	force_sig_info(SIGFPE, &info, task);
828 }
829 
830 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
831 {
832 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
833 	math_error(regs, error_code, X86_TRAP_MF);
834 }
835 
836 dotraplinkage void
837 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
838 {
839 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
840 	math_error(regs, error_code, X86_TRAP_XF);
841 }
842 
843 dotraplinkage void
844 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
845 {
846 	cond_local_irq_enable(regs);
847 }
848 
849 dotraplinkage void
850 do_device_not_available(struct pt_regs *regs, long error_code)
851 {
852 	unsigned long cr0;
853 
854 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
855 
856 #ifdef CONFIG_MATH_EMULATION
857 	if (!boot_cpu_has(X86_FEATURE_FPU) && (read_cr0() & X86_CR0_EM)) {
858 		struct math_emu_info info = { };
859 
860 		cond_local_irq_enable(regs);
861 
862 		info.regs = regs;
863 		math_emulate(&info);
864 		return;
865 	}
866 #endif
867 
868 	/* This should not happen. */
869 	cr0 = read_cr0();
870 	if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
871 		/* Try to fix it up and carry on. */
872 		write_cr0(cr0 & ~X86_CR0_TS);
873 	} else {
874 		/*
875 		 * Something terrible happened, and we're better off trying
876 		 * to kill the task than getting stuck in a never-ending
877 		 * loop of #NM faults.
878 		 */
879 		die("unexpected #NM exception", regs, error_code);
880 	}
881 }
882 NOKPROBE_SYMBOL(do_device_not_available);
883 
884 #ifdef CONFIG_X86_32
885 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
886 {
887 	siginfo_t info;
888 
889 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
890 	local_irq_enable();
891 
892 	info.si_signo = SIGILL;
893 	info.si_errno = 0;
894 	info.si_code = ILL_BADSTK;
895 	info.si_addr = NULL;
896 	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
897 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
898 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
899 			&info);
900 	}
901 }
902 #endif
903 
904 /* Set of traps needed for early debugging. */
905 void __init early_trap_init(void)
906 {
907 	/*
908 	 * Don't use IST to set DEBUG_STACK as it doesn't work until TSS
909 	 * is ready in cpu_init() <-- trap_init(). Before trap_init(),
910 	 * CPU runs at ring 0 so it is impossible to hit an invalid
911 	 * stack.  Using the original stack works well enough at this
912 	 * early stage. DEBUG_STACK will be equipped after cpu_init() in
913 	 * trap_init().
914 	 *
915 	 * We don't need to set trace_idt_table like set_intr_gate(),
916 	 * since we don't have trace_debug and it will be reset to
917 	 * 'debug' in trap_init() by set_intr_gate_ist().
918 	 */
919 	set_intr_gate_notrace(X86_TRAP_DB, debug);
920 	/* int3 can be called from all */
921 	set_system_intr_gate(X86_TRAP_BP, &int3);
922 #ifdef CONFIG_X86_32
923 	set_intr_gate(X86_TRAP_PF, page_fault);
924 #endif
925 	load_idt(&idt_descr);
926 }
927 
928 void __init early_trap_pf_init(void)
929 {
930 #ifdef CONFIG_X86_64
931 	set_intr_gate(X86_TRAP_PF, page_fault);
932 #endif
933 }
934 
935 void __init trap_init(void)
936 {
937 	int i;
938 
939 #ifdef CONFIG_EISA
940 	void __iomem *p = early_ioremap(0x0FFFD9, 4);
941 
942 	if (readl(p) == 'E' + ('I'<<8) + ('S'<<16) + ('A'<<24))
943 		EISA_bus = 1;
944 	early_iounmap(p, 4);
945 #endif
946 
947 	set_intr_gate(X86_TRAP_DE, divide_error);
948 	set_intr_gate_ist(X86_TRAP_NMI, &nmi, NMI_STACK);
949 	/* int4 can be called from all */
950 	set_system_intr_gate(X86_TRAP_OF, &overflow);
951 	set_intr_gate(X86_TRAP_BR, bounds);
952 	set_intr_gate(X86_TRAP_UD, invalid_op);
953 	set_intr_gate(X86_TRAP_NM, device_not_available);
954 #ifdef CONFIG_X86_32
955 	set_task_gate(X86_TRAP_DF, GDT_ENTRY_DOUBLEFAULT_TSS);
956 #else
957 	set_intr_gate_ist(X86_TRAP_DF, &double_fault, DOUBLEFAULT_STACK);
958 #endif
959 	set_intr_gate(X86_TRAP_OLD_MF, coprocessor_segment_overrun);
960 	set_intr_gate(X86_TRAP_TS, invalid_TSS);
961 	set_intr_gate(X86_TRAP_NP, segment_not_present);
962 	set_intr_gate(X86_TRAP_SS, stack_segment);
963 	set_intr_gate(X86_TRAP_GP, general_protection);
964 	set_intr_gate(X86_TRAP_SPURIOUS, spurious_interrupt_bug);
965 	set_intr_gate(X86_TRAP_MF, coprocessor_error);
966 	set_intr_gate(X86_TRAP_AC, alignment_check);
967 #ifdef CONFIG_X86_MCE
968 	set_intr_gate_ist(X86_TRAP_MC, &machine_check, MCE_STACK);
969 #endif
970 	set_intr_gate(X86_TRAP_XF, simd_coprocessor_error);
971 
972 	/* Reserve all the builtin and the syscall vector: */
973 	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++)
974 		set_bit(i, used_vectors);
975 
976 #ifdef CONFIG_IA32_EMULATION
977 	set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_compat);
978 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
979 #endif
980 
981 #ifdef CONFIG_X86_32
982 	set_system_intr_gate(IA32_SYSCALL_VECTOR, entry_INT80_32);
983 	set_bit(IA32_SYSCALL_VECTOR, used_vectors);
984 #endif
985 
986 	/*
987 	 * Set the IDT descriptor to a fixed read-only location, so that the
988 	 * "sidt" instruction will not leak the location of the kernel, and
989 	 * to defend the IDT against arbitrary memory write vulnerabilities.
990 	 * It will be reloaded in cpu_init() */
991 	__set_fixmap(FIX_RO_IDT, __pa_symbol(idt_table), PAGE_KERNEL_RO);
992 	idt_descr.address = fix_to_virt(FIX_RO_IDT);
993 
994 	/*
995 	 * Should be a barrier for any external CPU state:
996 	 */
997 	cpu_init();
998 
999 	/*
1000 	 * X86_TRAP_DB and X86_TRAP_BP have been set
1001 	 * in early_trap_init(). However, ITS works only after
1002 	 * cpu_init() loads TSS. See comments in early_trap_init().
1003 	 */
1004 	set_intr_gate_ist(X86_TRAP_DB, &debug, DEBUG_STACK);
1005 	/* int3 can be called from all */
1006 	set_system_intr_gate_ist(X86_TRAP_BP, &int3, DEBUG_STACK);
1007 
1008 	x86_init.irqs.trap_init();
1009 
1010 #ifdef CONFIG_X86_64
1011 	memcpy(&debug_idt_table, &idt_table, IDT_ENTRIES * 16);
1012 	set_nmi_gate(X86_TRAP_DB, &debug);
1013 	set_nmi_gate(X86_TRAP_BP, &int3);
1014 #endif
1015 }
1016