xref: /linux/arch/x86/kernel/traps.c (revision 63307d015b91e626c97bb82e88054af3d0b74643)
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 #if defined(CONFIG_EDAC)
42 #include <linux/edac.h>
43 #endif
44 
45 #include <asm/stacktrace.h>
46 #include <asm/processor.h>
47 #include <asm/debugreg.h>
48 #include <linux/atomic.h>
49 #include <asm/text-patching.h>
50 #include <asm/ftrace.h>
51 #include <asm/traps.h>
52 #include <asm/desc.h>
53 #include <asm/fpu/internal.h>
54 #include <asm/cpu_entry_area.h>
55 #include <asm/mce.h>
56 #include <asm/fixmap.h>
57 #include <asm/mach_traps.h>
58 #include <asm/alternative.h>
59 #include <asm/fpu/xstate.h>
60 #include <asm/trace/mpx.h>
61 #include <asm/mpx.h>
62 #include <asm/vm86.h>
63 #include <asm/umip.h>
64 
65 #ifdef CONFIG_X86_64
66 #include <asm/x86_init.h>
67 #include <asm/pgalloc.h>
68 #include <asm/proto.h>
69 #else
70 #include <asm/processor-flags.h>
71 #include <asm/setup.h>
72 #include <asm/proto.h>
73 #endif
74 
75 DECLARE_BITMAP(system_vectors, NR_VECTORS);
76 
77 static inline void cond_local_irq_enable(struct pt_regs *regs)
78 {
79 	if (regs->flags & X86_EFLAGS_IF)
80 		local_irq_enable();
81 }
82 
83 static inline void cond_local_irq_disable(struct pt_regs *regs)
84 {
85 	if (regs->flags & X86_EFLAGS_IF)
86 		local_irq_disable();
87 }
88 
89 /*
90  * In IST context, we explicitly disable preemption.  This serves two
91  * purposes: it makes it much less likely that we would accidentally
92  * schedule in IST context and it will force a warning if we somehow
93  * manage to schedule by accident.
94  */
95 void ist_enter(struct pt_regs *regs)
96 {
97 	if (user_mode(regs)) {
98 		RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
99 	} else {
100 		/*
101 		 * We might have interrupted pretty much anything.  In
102 		 * fact, if we're a machine check, we can even interrupt
103 		 * NMI processing.  We don't want in_nmi() to return true,
104 		 * but we need to notify RCU.
105 		 */
106 		rcu_nmi_enter();
107 	}
108 
109 	preempt_disable();
110 
111 	/* This code is a bit fragile.  Test it. */
112 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "ist_enter didn't work");
113 }
114 NOKPROBE_SYMBOL(ist_enter);
115 
116 void ist_exit(struct pt_regs *regs)
117 {
118 	preempt_enable_no_resched();
119 
120 	if (!user_mode(regs))
121 		rcu_nmi_exit();
122 }
123 
124 /**
125  * ist_begin_non_atomic() - begin a non-atomic section in an IST exception
126  * @regs:	regs passed to the IST exception handler
127  *
128  * IST exception handlers normally cannot schedule.  As a special
129  * exception, if the exception interrupted userspace code (i.e.
130  * user_mode(regs) would return true) and the exception was not
131  * a double fault, it can be safe to schedule.  ist_begin_non_atomic()
132  * begins a non-atomic section within an ist_enter()/ist_exit() region.
133  * Callers are responsible for enabling interrupts themselves inside
134  * the non-atomic section, and callers must call ist_end_non_atomic()
135  * before ist_exit().
136  */
137 void ist_begin_non_atomic(struct pt_regs *regs)
138 {
139 	BUG_ON(!user_mode(regs));
140 
141 	/*
142 	 * Sanity check: we need to be on the normal thread stack.  This
143 	 * will catch asm bugs and any attempt to use ist_preempt_enable
144 	 * from double_fault.
145 	 */
146 	BUG_ON(!on_thread_stack());
147 
148 	preempt_enable_no_resched();
149 }
150 
151 /**
152  * ist_end_non_atomic() - begin a non-atomic section in an IST exception
153  *
154  * Ends a non-atomic section started with ist_begin_non_atomic().
155  */
156 void ist_end_non_atomic(void)
157 {
158 	preempt_disable();
159 }
160 
161 int is_valid_bugaddr(unsigned long addr)
162 {
163 	unsigned short ud;
164 
165 	if (addr < TASK_SIZE_MAX)
166 		return 0;
167 
168 	if (probe_kernel_address((unsigned short *)addr, ud))
169 		return 0;
170 
171 	return ud == INSN_UD0 || ud == INSN_UD2;
172 }
173 
174 int fixup_bug(struct pt_regs *regs, int trapnr)
175 {
176 	if (trapnr != X86_TRAP_UD)
177 		return 0;
178 
179 	switch (report_bug(regs->ip, regs)) {
180 	case BUG_TRAP_TYPE_NONE:
181 	case BUG_TRAP_TYPE_BUG:
182 		break;
183 
184 	case BUG_TRAP_TYPE_WARN:
185 		regs->ip += LEN_UD2;
186 		return 1;
187 	}
188 
189 	return 0;
190 }
191 
192 static nokprobe_inline int
193 do_trap_no_signal(struct task_struct *tsk, int trapnr, const char *str,
194 		  struct pt_regs *regs,	long error_code)
195 {
196 	if (v8086_mode(regs)) {
197 		/*
198 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
199 		 * On nmi (interrupt 2), do_trap should not be called.
200 		 */
201 		if (trapnr < X86_TRAP_UD) {
202 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
203 						error_code, trapnr))
204 				return 0;
205 		}
206 	} else if (!user_mode(regs)) {
207 		if (fixup_exception(regs, trapnr, error_code, 0))
208 			return 0;
209 
210 		tsk->thread.error_code = error_code;
211 		tsk->thread.trap_nr = trapnr;
212 		die(str, regs, error_code);
213 	}
214 
215 	/*
216 	 * We want error_code and trap_nr set for userspace faults and
217 	 * kernelspace faults which result in die(), but not
218 	 * kernelspace faults which are fixed up.  die() gives the
219 	 * process no chance to handle the signal and notice the
220 	 * kernel fault information, so that won't result in polluting
221 	 * the information about previously queued, but not yet
222 	 * delivered, faults.  See also do_general_protection below.
223 	 */
224 	tsk->thread.error_code = error_code;
225 	tsk->thread.trap_nr = trapnr;
226 
227 	return -1;
228 }
229 
230 static void show_signal(struct task_struct *tsk, int signr,
231 			const char *type, const char *desc,
232 			struct pt_regs *regs, long error_code)
233 {
234 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
235 	    printk_ratelimit()) {
236 		pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
237 			tsk->comm, task_pid_nr(tsk), type, desc,
238 			regs->ip, regs->sp, error_code);
239 		print_vma_addr(KERN_CONT " in ", regs->ip);
240 		pr_cont("\n");
241 	}
242 }
243 
244 static void
245 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
246 	long error_code, int sicode, void __user *addr)
247 {
248 	struct task_struct *tsk = current;
249 
250 
251 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
252 		return;
253 
254 	show_signal(tsk, signr, "trap ", str, regs, error_code);
255 
256 	if (!sicode)
257 		force_sig(signr, tsk);
258 	else
259 		force_sig_fault(signr, sicode, addr, tsk);
260 }
261 NOKPROBE_SYMBOL(do_trap);
262 
263 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
264 	unsigned long trapnr, int signr, int sicode, void __user *addr)
265 {
266 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
267 
268 	/*
269 	 * WARN*()s end up here; fix them up before we call the
270 	 * notifier chain.
271 	 */
272 	if (!user_mode(regs) && fixup_bug(regs, trapnr))
273 		return;
274 
275 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
276 			NOTIFY_STOP) {
277 		cond_local_irq_enable(regs);
278 		do_trap(trapnr, signr, str, regs, error_code, sicode, addr);
279 	}
280 }
281 
282 #define IP ((void __user *)uprobe_get_trap_addr(regs))
283 #define DO_ERROR(trapnr, signr, sicode, addr, str, name)		   \
284 dotraplinkage void do_##name(struct pt_regs *regs, long error_code)	   \
285 {									   \
286 	do_error_trap(regs, error_code, str, trapnr, signr, sicode, addr); \
287 }
288 
289 DO_ERROR(X86_TRAP_DE,     SIGFPE,  FPE_INTDIV,   IP, "divide error",        divide_error)
290 DO_ERROR(X86_TRAP_OF,     SIGSEGV,          0, NULL, "overflow",            overflow)
291 DO_ERROR(X86_TRAP_UD,     SIGILL,  ILL_ILLOPN,   IP, "invalid opcode",      invalid_op)
292 DO_ERROR(X86_TRAP_OLD_MF, SIGFPE,           0, NULL, "coprocessor segment overrun", coprocessor_segment_overrun)
293 DO_ERROR(X86_TRAP_TS,     SIGSEGV,          0, NULL, "invalid TSS",         invalid_TSS)
294 DO_ERROR(X86_TRAP_NP,     SIGBUS,           0, NULL, "segment not present", segment_not_present)
295 DO_ERROR(X86_TRAP_SS,     SIGBUS,           0, NULL, "stack segment",       stack_segment)
296 DO_ERROR(X86_TRAP_AC,     SIGBUS,  BUS_ADRALN, NULL, "alignment check",     alignment_check)
297 #undef IP
298 
299 #ifdef CONFIG_VMAP_STACK
300 __visible void __noreturn handle_stack_overflow(const char *message,
301 						struct pt_regs *regs,
302 						unsigned long fault_address)
303 {
304 	printk(KERN_EMERG "BUG: stack guard page was hit at %p (stack is %p..%p)\n",
305 		 (void *)fault_address, current->stack,
306 		 (char *)current->stack + THREAD_SIZE - 1);
307 	die(message, regs, 0);
308 
309 	/* Be absolutely certain we don't return. */
310 	panic("%s", message);
311 }
312 #endif
313 
314 #ifdef CONFIG_X86_64
315 /* Runs on IST stack */
316 dotraplinkage void do_double_fault(struct pt_regs *regs, long error_code)
317 {
318 	static const char str[] = "double fault";
319 	struct task_struct *tsk = current;
320 #ifdef CONFIG_VMAP_STACK
321 	unsigned long cr2;
322 #endif
323 
324 #ifdef CONFIG_X86_ESPFIX64
325 	extern unsigned char native_irq_return_iret[];
326 
327 	/*
328 	 * If IRET takes a non-IST fault on the espfix64 stack, then we
329 	 * end up promoting it to a doublefault.  In that case, take
330 	 * advantage of the fact that we're not using the normal (TSS.sp0)
331 	 * stack right now.  We can write a fake #GP(0) frame at TSS.sp0
332 	 * and then modify our own IRET frame so that, when we return,
333 	 * we land directly at the #GP(0) vector with the stack already
334 	 * set up according to its expectations.
335 	 *
336 	 * The net result is that our #GP handler will think that we
337 	 * entered from usermode with the bad user context.
338 	 *
339 	 * No need for ist_enter here because we don't use RCU.
340 	 */
341 	if (((long)regs->sp >> P4D_SHIFT) == ESPFIX_PGD_ENTRY &&
342 		regs->cs == __KERNEL_CS &&
343 		regs->ip == (unsigned long)native_irq_return_iret)
344 	{
345 		struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
346 
347 		/*
348 		 * regs->sp points to the failing IRET frame on the
349 		 * ESPFIX64 stack.  Copy it to the entry stack.  This fills
350 		 * in gpregs->ss through gpregs->ip.
351 		 *
352 		 */
353 		memmove(&gpregs->ip, (void *)regs->sp, 5*8);
354 		gpregs->orig_ax = 0;  /* Missing (lost) #GP error code */
355 
356 		/*
357 		 * Adjust our frame so that we return straight to the #GP
358 		 * vector with the expected RSP value.  This is safe because
359 		 * we won't enable interupts or schedule before we invoke
360 		 * general_protection, so nothing will clobber the stack
361 		 * frame we just set up.
362 		 *
363 		 * We will enter general_protection with kernel GSBASE,
364 		 * which is what the stub expects, given that the faulting
365 		 * RIP will be the IRET instruction.
366 		 */
367 		regs->ip = (unsigned long)general_protection;
368 		regs->sp = (unsigned long)&gpregs->orig_ax;
369 
370 		return;
371 	}
372 #endif
373 
374 	ist_enter(regs);
375 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
376 
377 	tsk->thread.error_code = error_code;
378 	tsk->thread.trap_nr = X86_TRAP_DF;
379 
380 #ifdef CONFIG_VMAP_STACK
381 	/*
382 	 * If we overflow the stack into a guard page, the CPU will fail
383 	 * to deliver #PF and will send #DF instead.  Similarly, if we
384 	 * take any non-IST exception while too close to the bottom of
385 	 * the stack, the processor will get a page fault while
386 	 * delivering the exception and will generate a double fault.
387 	 *
388 	 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
389 	 * Page-Fault Exception (#PF):
390 	 *
391 	 *   Processors update CR2 whenever a page fault is detected. If a
392 	 *   second page fault occurs while an earlier page fault is being
393 	 *   delivered, the faulting linear address of the second fault will
394 	 *   overwrite the contents of CR2 (replacing the previous
395 	 *   address). These updates to CR2 occur even if the page fault
396 	 *   results in a double fault or occurs during the delivery of a
397 	 *   double fault.
398 	 *
399 	 * The logic below has a small possibility of incorrectly diagnosing
400 	 * some errors as stack overflows.  For example, if the IDT or GDT
401 	 * gets corrupted such that #GP delivery fails due to a bad descriptor
402 	 * causing #GP and we hit this condition while CR2 coincidentally
403 	 * points to the stack guard page, we'll think we overflowed the
404 	 * stack.  Given that we're going to panic one way or another
405 	 * if this happens, this isn't necessarily worth fixing.
406 	 *
407 	 * If necessary, we could improve the test by only diagnosing
408 	 * a stack overflow if the saved RSP points within 47 bytes of
409 	 * the bottom of the stack: if RSP == tsk_stack + 48 and we
410 	 * take an exception, the stack is already aligned and there
411 	 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
412 	 * possible error code, so a stack overflow would *not* double
413 	 * fault.  With any less space left, exception delivery could
414 	 * fail, and, as a practical matter, we've overflowed the
415 	 * stack even if the actual trigger for the double fault was
416 	 * something else.
417 	 */
418 	cr2 = read_cr2();
419 	if ((unsigned long)task_stack_page(tsk) - 1 - cr2 < PAGE_SIZE)
420 		handle_stack_overflow("kernel stack overflow (double-fault)", regs, cr2);
421 #endif
422 
423 #ifdef CONFIG_DOUBLEFAULT
424 	df_debug(regs, error_code);
425 #endif
426 	/*
427 	 * This is always a kernel trap and never fixable (and thus must
428 	 * never return).
429 	 */
430 	for (;;)
431 		die(str, regs, error_code);
432 }
433 #endif
434 
435 dotraplinkage void do_bounds(struct pt_regs *regs, long error_code)
436 {
437 	const struct mpx_bndcsr *bndcsr;
438 
439 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
440 	if (notify_die(DIE_TRAP, "bounds", regs, error_code,
441 			X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
442 		return;
443 	cond_local_irq_enable(regs);
444 
445 	if (!user_mode(regs))
446 		die("bounds", regs, error_code);
447 
448 	if (!cpu_feature_enabled(X86_FEATURE_MPX)) {
449 		/* The exception is not from Intel MPX */
450 		goto exit_trap;
451 	}
452 
453 	/*
454 	 * We need to look at BNDSTATUS to resolve this exception.
455 	 * A NULL here might mean that it is in its 'init state',
456 	 * which is all zeros which indicates MPX was not
457 	 * responsible for the exception.
458 	 */
459 	bndcsr = get_xsave_field_ptr(XFEATURE_BNDCSR);
460 	if (!bndcsr)
461 		goto exit_trap;
462 
463 	trace_bounds_exception_mpx(bndcsr);
464 	/*
465 	 * The error code field of the BNDSTATUS register communicates status
466 	 * information of a bound range exception #BR or operation involving
467 	 * bound directory.
468 	 */
469 	switch (bndcsr->bndstatus & MPX_BNDSTA_ERROR_CODE) {
470 	case 2:	/* Bound directory has invalid entry. */
471 		if (mpx_handle_bd_fault())
472 			goto exit_trap;
473 		break; /* Success, it was handled */
474 	case 1: /* Bound violation. */
475 	{
476 		struct task_struct *tsk = current;
477 		struct mpx_fault_info mpx;
478 
479 		if (mpx_fault_info(&mpx, regs)) {
480 			/*
481 			 * We failed to decode the MPX instruction.  Act as if
482 			 * the exception was not caused by MPX.
483 			 */
484 			goto exit_trap;
485 		}
486 		/*
487 		 * Success, we decoded the instruction and retrieved
488 		 * an 'mpx' containing the address being accessed
489 		 * which caused the exception.  This information
490 		 * allows and application to possibly handle the
491 		 * #BR exception itself.
492 		 */
493 		if (!do_trap_no_signal(tsk, X86_TRAP_BR, "bounds", regs,
494 				       error_code))
495 			break;
496 
497 		show_signal(tsk, SIGSEGV, "trap ", "bounds", regs, error_code);
498 
499 		force_sig_bnderr(mpx.addr, mpx.lower, mpx.upper);
500 		break;
501 	}
502 	case 0: /* No exception caused by Intel MPX operations. */
503 		goto exit_trap;
504 	default:
505 		die("bounds", regs, error_code);
506 	}
507 
508 	return;
509 
510 exit_trap:
511 	/*
512 	 * This path out is for all the cases where we could not
513 	 * handle the exception in some way (like allocating a
514 	 * table or telling userspace about it.  We will also end
515 	 * up here if the kernel has MPX turned off at compile
516 	 * time..
517 	 */
518 	do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, error_code, 0, NULL);
519 }
520 
521 dotraplinkage void
522 do_general_protection(struct pt_regs *regs, long error_code)
523 {
524 	const char *desc = "general protection fault";
525 	struct task_struct *tsk;
526 
527 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
528 	cond_local_irq_enable(regs);
529 
530 	if (static_cpu_has(X86_FEATURE_UMIP)) {
531 		if (user_mode(regs) && fixup_umip_exception(regs))
532 			return;
533 	}
534 
535 	if (v8086_mode(regs)) {
536 		local_irq_enable();
537 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
538 		return;
539 	}
540 
541 	tsk = current;
542 	if (!user_mode(regs)) {
543 		if (fixup_exception(regs, X86_TRAP_GP, error_code, 0))
544 			return;
545 
546 		tsk->thread.error_code = error_code;
547 		tsk->thread.trap_nr = X86_TRAP_GP;
548 
549 		/*
550 		 * To be potentially processing a kprobe fault and to
551 		 * trust the result from kprobe_running(), we have to
552 		 * be non-preemptible.
553 		 */
554 		if (!preemptible() && kprobe_running() &&
555 		    kprobe_fault_handler(regs, X86_TRAP_GP))
556 			return;
557 
558 		if (notify_die(DIE_GPF, desc, regs, error_code,
559 			       X86_TRAP_GP, SIGSEGV) != NOTIFY_STOP)
560 			die(desc, regs, error_code);
561 		return;
562 	}
563 
564 	tsk->thread.error_code = error_code;
565 	tsk->thread.trap_nr = X86_TRAP_GP;
566 
567 	show_signal(tsk, SIGSEGV, "", desc, regs, error_code);
568 
569 	force_sig(SIGSEGV, tsk);
570 }
571 NOKPROBE_SYMBOL(do_general_protection);
572 
573 dotraplinkage void notrace do_int3(struct pt_regs *regs, long error_code)
574 {
575 #ifdef CONFIG_DYNAMIC_FTRACE
576 	/*
577 	 * ftrace must be first, everything else may cause a recursive crash.
578 	 * See note by declaration of modifying_ftrace_code in ftrace.c
579 	 */
580 	if (unlikely(atomic_read(&modifying_ftrace_code)) &&
581 	    ftrace_int3_handler(regs))
582 		return;
583 #endif
584 	if (poke_int3_handler(regs))
585 		return;
586 
587 	/*
588 	 * Use ist_enter despite the fact that we don't use an IST stack.
589 	 * We can be called from a kprobe in non-CONTEXT_KERNEL kernel
590 	 * mode or even during context tracking state changes.
591 	 *
592 	 * This means that we can't schedule.  That's okay.
593 	 */
594 	ist_enter(regs);
595 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
596 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
597 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
598 				SIGTRAP) == NOTIFY_STOP)
599 		goto exit;
600 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
601 
602 #ifdef CONFIG_KPROBES
603 	if (kprobe_int3_handler(regs))
604 		goto exit;
605 #endif
606 
607 	if (notify_die(DIE_INT3, "int3", regs, error_code, X86_TRAP_BP,
608 			SIGTRAP) == NOTIFY_STOP)
609 		goto exit;
610 
611 	cond_local_irq_enable(regs);
612 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, error_code, 0, NULL);
613 	cond_local_irq_disable(regs);
614 
615 exit:
616 	ist_exit(regs);
617 }
618 NOKPROBE_SYMBOL(do_int3);
619 
620 #ifdef CONFIG_X86_64
621 /*
622  * Help handler running on a per-cpu (IST or entry trampoline) stack
623  * to switch to the normal thread stack if the interrupted code was in
624  * user mode. The actual stack switch is done in entry_64.S
625  */
626 asmlinkage __visible notrace struct pt_regs *sync_regs(struct pt_regs *eregs)
627 {
628 	struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
629 	if (regs != eregs)
630 		*regs = *eregs;
631 	return regs;
632 }
633 NOKPROBE_SYMBOL(sync_regs);
634 
635 struct bad_iret_stack {
636 	void *error_entry_ret;
637 	struct pt_regs regs;
638 };
639 
640 asmlinkage __visible notrace
641 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
642 {
643 	/*
644 	 * This is called from entry_64.S early in handling a fault
645 	 * caused by a bad iret to user mode.  To handle the fault
646 	 * correctly, we want to move our stack frame to where it would
647 	 * be had we entered directly on the entry stack (rather than
648 	 * just below the IRET frame) and we want to pretend that the
649 	 * exception came from the IRET target.
650 	 */
651 	struct bad_iret_stack *new_stack =
652 		(struct bad_iret_stack *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
653 
654 	/* Copy the IRET target to the new stack. */
655 	memmove(&new_stack->regs.ip, (void *)s->regs.sp, 5*8);
656 
657 	/* Copy the remainder of the stack from the current stack. */
658 	memmove(new_stack, s, offsetof(struct bad_iret_stack, regs.ip));
659 
660 	BUG_ON(!user_mode(&new_stack->regs));
661 	return new_stack;
662 }
663 NOKPROBE_SYMBOL(fixup_bad_iret);
664 #endif
665 
666 static bool is_sysenter_singlestep(struct pt_regs *regs)
667 {
668 	/*
669 	 * We don't try for precision here.  If we're anywhere in the region of
670 	 * code that can be single-stepped in the SYSENTER entry path, then
671 	 * assume that this is a useless single-step trap due to SYSENTER
672 	 * being invoked with TF set.  (We don't know in advance exactly
673 	 * which instructions will be hit because BTF could plausibly
674 	 * be set.)
675 	 */
676 #ifdef CONFIG_X86_32
677 	return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
678 		(unsigned long)__end_SYSENTER_singlestep_region -
679 		(unsigned long)__begin_SYSENTER_singlestep_region;
680 #elif defined(CONFIG_IA32_EMULATION)
681 	return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
682 		(unsigned long)__end_entry_SYSENTER_compat -
683 		(unsigned long)entry_SYSENTER_compat;
684 #else
685 	return false;
686 #endif
687 }
688 
689 /*
690  * Our handling of the processor debug registers is non-trivial.
691  * We do not clear them on entry and exit from the kernel. Therefore
692  * it is possible to get a watchpoint trap here from inside the kernel.
693  * However, the code in ./ptrace.c has ensured that the user can
694  * only set watchpoints on userspace addresses. Therefore the in-kernel
695  * watchpoint trap can only occur in code which is reading/writing
696  * from user space. Such code must not hold kernel locks (since it
697  * can equally take a page fault), therefore it is safe to call
698  * force_sig_info even though that claims and releases locks.
699  *
700  * Code in ./signal.c ensures that the debug control register
701  * is restored before we deliver any signal, and therefore that
702  * user code runs with the correct debug control register even though
703  * we clear it here.
704  *
705  * Being careful here means that we don't have to be as careful in a
706  * lot of more complicated places (task switching can be a bit lazy
707  * about restoring all the debug state, and ptrace doesn't have to
708  * find every occurrence of the TF bit that could be saved away even
709  * by user code)
710  *
711  * May run on IST stack.
712  */
713 dotraplinkage void do_debug(struct pt_regs *regs, long error_code)
714 {
715 	struct task_struct *tsk = current;
716 	int user_icebp = 0;
717 	unsigned long dr6;
718 	int si_code;
719 
720 	ist_enter(regs);
721 
722 	get_debugreg(dr6, 6);
723 	/*
724 	 * The Intel SDM says:
725 	 *
726 	 *   Certain debug exceptions may clear bits 0-3. The remaining
727 	 *   contents of the DR6 register are never cleared by the
728 	 *   processor. To avoid confusion in identifying debug
729 	 *   exceptions, debug handlers should clear the register before
730 	 *   returning to the interrupted task.
731 	 *
732 	 * Keep it simple: clear DR6 immediately.
733 	 */
734 	set_debugreg(0, 6);
735 
736 	/* Filter out all the reserved bits which are preset to 1 */
737 	dr6 &= ~DR6_RESERVED;
738 
739 	/*
740 	 * The SDM says "The processor clears the BTF flag when it
741 	 * generates a debug exception."  Clear TIF_BLOCKSTEP to keep
742 	 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
743 	 */
744 	clear_tsk_thread_flag(tsk, TIF_BLOCKSTEP);
745 
746 	if (unlikely(!user_mode(regs) && (dr6 & DR_STEP) &&
747 		     is_sysenter_singlestep(regs))) {
748 		dr6 &= ~DR_STEP;
749 		if (!dr6)
750 			goto exit;
751 		/*
752 		 * else we might have gotten a single-step trap and hit a
753 		 * watchpoint at the same time, in which case we should fall
754 		 * through and handle the watchpoint.
755 		 */
756 	}
757 
758 	/*
759 	 * If dr6 has no reason to give us about the origin of this trap,
760 	 * then it's very likely the result of an icebp/int01 trap.
761 	 * User wants a sigtrap for that.
762 	 */
763 	if (!dr6 && user_mode(regs))
764 		user_icebp = 1;
765 
766 	/* Store the virtualized DR6 value */
767 	tsk->thread.debugreg6 = dr6;
768 
769 #ifdef CONFIG_KPROBES
770 	if (kprobe_debug_handler(regs))
771 		goto exit;
772 #endif
773 
774 	if (notify_die(DIE_DEBUG, "debug", regs, (long)&dr6, error_code,
775 							SIGTRAP) == NOTIFY_STOP)
776 		goto exit;
777 
778 	/*
779 	 * Let others (NMI) know that the debug stack is in use
780 	 * as we may switch to the interrupt stack.
781 	 */
782 	debug_stack_usage_inc();
783 
784 	/* It's safe to allow irq's after DR6 has been saved */
785 	cond_local_irq_enable(regs);
786 
787 	if (v8086_mode(regs)) {
788 		handle_vm86_trap((struct kernel_vm86_regs *) regs, error_code,
789 					X86_TRAP_DB);
790 		cond_local_irq_disable(regs);
791 		debug_stack_usage_dec();
792 		goto exit;
793 	}
794 
795 	if (WARN_ON_ONCE((dr6 & DR_STEP) && !user_mode(regs))) {
796 		/*
797 		 * Historical junk that used to handle SYSENTER single-stepping.
798 		 * This should be unreachable now.  If we survive for a while
799 		 * without anyone hitting this warning, we'll turn this into
800 		 * an oops.
801 		 */
802 		tsk->thread.debugreg6 &= ~DR_STEP;
803 		set_tsk_thread_flag(tsk, TIF_SINGLESTEP);
804 		regs->flags &= ~X86_EFLAGS_TF;
805 	}
806 	si_code = get_si_code(tsk->thread.debugreg6);
807 	if (tsk->thread.debugreg6 & (DR_STEP | DR_TRAP_BITS) || user_icebp)
808 		send_sigtrap(tsk, regs, error_code, si_code);
809 	cond_local_irq_disable(regs);
810 	debug_stack_usage_dec();
811 
812 exit:
813 	ist_exit(regs);
814 }
815 NOKPROBE_SYMBOL(do_debug);
816 
817 /*
818  * Note that we play around with the 'TS' bit in an attempt to get
819  * the correct behaviour even in the presence of the asynchronous
820  * IRQ13 behaviour
821  */
822 static void math_error(struct pt_regs *regs, int error_code, int trapnr)
823 {
824 	struct task_struct *task = current;
825 	struct fpu *fpu = &task->thread.fpu;
826 	int si_code;
827 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
828 						"simd exception";
829 
830 	cond_local_irq_enable(regs);
831 
832 	if (!user_mode(regs)) {
833 		if (fixup_exception(regs, trapnr, error_code, 0))
834 			return;
835 
836 		task->thread.error_code = error_code;
837 		task->thread.trap_nr = trapnr;
838 
839 		if (notify_die(DIE_TRAP, str, regs, error_code,
840 					trapnr, SIGFPE) != NOTIFY_STOP)
841 			die(str, regs, error_code);
842 		return;
843 	}
844 
845 	/*
846 	 * Save the info for the exception handler and clear the error.
847 	 */
848 	fpu__save(fpu);
849 
850 	task->thread.trap_nr	= trapnr;
851 	task->thread.error_code = error_code;
852 
853 	si_code = fpu__exception_code(fpu, trapnr);
854 	/* Retry when we get spurious exceptions: */
855 	if (!si_code)
856 		return;
857 
858 	force_sig_fault(SIGFPE, si_code,
859 			(void __user *)uprobe_get_trap_addr(regs), task);
860 }
861 
862 dotraplinkage void do_coprocessor_error(struct pt_regs *regs, long error_code)
863 {
864 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
865 	math_error(regs, error_code, X86_TRAP_MF);
866 }
867 
868 dotraplinkage void
869 do_simd_coprocessor_error(struct pt_regs *regs, long error_code)
870 {
871 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
872 	math_error(regs, error_code, X86_TRAP_XF);
873 }
874 
875 dotraplinkage void
876 do_spurious_interrupt_bug(struct pt_regs *regs, long error_code)
877 {
878 	cond_local_irq_enable(regs);
879 }
880 
881 dotraplinkage void
882 do_device_not_available(struct pt_regs *regs, long error_code)
883 {
884 	unsigned long cr0 = read_cr0();
885 
886 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
887 
888 #ifdef CONFIG_MATH_EMULATION
889 	if (!boot_cpu_has(X86_FEATURE_FPU) && (cr0 & X86_CR0_EM)) {
890 		struct math_emu_info info = { };
891 
892 		cond_local_irq_enable(regs);
893 
894 		info.regs = regs;
895 		math_emulate(&info);
896 		return;
897 	}
898 #endif
899 
900 	/* This should not happen. */
901 	if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
902 		/* Try to fix it up and carry on. */
903 		write_cr0(cr0 & ~X86_CR0_TS);
904 	} else {
905 		/*
906 		 * Something terrible happened, and we're better off trying
907 		 * to kill the task than getting stuck in a never-ending
908 		 * loop of #NM faults.
909 		 */
910 		die("unexpected #NM exception", regs, error_code);
911 	}
912 }
913 NOKPROBE_SYMBOL(do_device_not_available);
914 
915 #ifdef CONFIG_X86_32
916 dotraplinkage void do_iret_error(struct pt_regs *regs, long error_code)
917 {
918 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
919 	local_irq_enable();
920 
921 	if (notify_die(DIE_TRAP, "iret exception", regs, error_code,
922 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
923 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, error_code,
924 			ILL_BADSTK, (void __user *)NULL);
925 	}
926 }
927 #endif
928 
929 void __init trap_init(void)
930 {
931 	/* Init cpu_entry_area before IST entries are set up */
932 	setup_cpu_entry_areas();
933 
934 	idt_setup_traps();
935 
936 	/*
937 	 * Set the IDT descriptor to a fixed read-only location, so that the
938 	 * "sidt" instruction will not leak the location of the kernel, and
939 	 * to defend the IDT against arbitrary memory write vulnerabilities.
940 	 * It will be reloaded in cpu_init() */
941 	cea_set_pte(CPU_ENTRY_AREA_RO_IDT_VADDR, __pa_symbol(idt_table),
942 		    PAGE_KERNEL_RO);
943 	idt_descr.address = CPU_ENTRY_AREA_RO_IDT;
944 
945 	/*
946 	 * Should be a barrier for any external CPU state:
947 	 */
948 	cpu_init();
949 
950 	idt_setup_ist_traps();
951 
952 	x86_init.irqs.trap_init();
953 
954 	idt_setup_debugidt_traps();
955 }
956