xref: /openbmc/linux/arch/x86/kernel/traps.c (revision 2208f39c)
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 #include <linux/hardirq.h>
41 #include <linux/atomic.h>
42 
43 #include <asm/stacktrace.h>
44 #include <asm/processor.h>
45 #include <asm/debugreg.h>
46 #include <asm/realmode.h>
47 #include <asm/text-patching.h>
48 #include <asm/ftrace.h>
49 #include <asm/traps.h>
50 #include <asm/desc.h>
51 #include <asm/fpu/internal.h>
52 #include <asm/cpu.h>
53 #include <asm/cpu_entry_area.h>
54 #include <asm/mce.h>
55 #include <asm/fixmap.h>
56 #include <asm/mach_traps.h>
57 #include <asm/alternative.h>
58 #include <asm/fpu/xstate.h>
59 #include <asm/vm86.h>
60 #include <asm/umip.h>
61 #include <asm/insn.h>
62 #include <asm/insn-eval.h>
63 
64 #ifdef CONFIG_X86_64
65 #include <asm/x86_init.h>
66 #include <asm/proto.h>
67 #else
68 #include <asm/processor-flags.h>
69 #include <asm/setup.h>
70 #include <asm/proto.h>
71 #endif
72 
73 DECLARE_BITMAP(system_vectors, NR_VECTORS);
74 
75 static inline void cond_local_irq_enable(struct pt_regs *regs)
76 {
77 	if (regs->flags & X86_EFLAGS_IF)
78 		local_irq_enable();
79 }
80 
81 static inline void cond_local_irq_disable(struct pt_regs *regs)
82 {
83 	if (regs->flags & X86_EFLAGS_IF)
84 		local_irq_disable();
85 }
86 
87 __always_inline int is_valid_bugaddr(unsigned long addr)
88 {
89 	if (addr < TASK_SIZE_MAX)
90 		return 0;
91 
92 	/*
93 	 * We got #UD, if the text isn't readable we'd have gotten
94 	 * a different exception.
95 	 */
96 	return *(unsigned short *)addr == INSN_UD2;
97 }
98 
99 static nokprobe_inline int
100 do_trap_no_signal(struct task_struct *tsk, int trapnr, const char *str,
101 		  struct pt_regs *regs,	long error_code)
102 {
103 	if (v8086_mode(regs)) {
104 		/*
105 		 * Traps 0, 1, 3, 4, and 5 should be forwarded to vm86.
106 		 * On nmi (interrupt 2), do_trap should not be called.
107 		 */
108 		if (trapnr < X86_TRAP_UD) {
109 			if (!handle_vm86_trap((struct kernel_vm86_regs *) regs,
110 						error_code, trapnr))
111 				return 0;
112 		}
113 	} else if (!user_mode(regs)) {
114 		if (fixup_exception(regs, trapnr, error_code, 0))
115 			return 0;
116 
117 		tsk->thread.error_code = error_code;
118 		tsk->thread.trap_nr = trapnr;
119 		die(str, regs, error_code);
120 	}
121 
122 	/*
123 	 * We want error_code and trap_nr set for userspace faults and
124 	 * kernelspace faults which result in die(), but not
125 	 * kernelspace faults which are fixed up.  die() gives the
126 	 * process no chance to handle the signal and notice the
127 	 * kernel fault information, so that won't result in polluting
128 	 * the information about previously queued, but not yet
129 	 * delivered, faults.  See also exc_general_protection below.
130 	 */
131 	tsk->thread.error_code = error_code;
132 	tsk->thread.trap_nr = trapnr;
133 
134 	return -1;
135 }
136 
137 static void show_signal(struct task_struct *tsk, int signr,
138 			const char *type, const char *desc,
139 			struct pt_regs *regs, long error_code)
140 {
141 	if (show_unhandled_signals && unhandled_signal(tsk, signr) &&
142 	    printk_ratelimit()) {
143 		pr_info("%s[%d] %s%s ip:%lx sp:%lx error:%lx",
144 			tsk->comm, task_pid_nr(tsk), type, desc,
145 			regs->ip, regs->sp, error_code);
146 		print_vma_addr(KERN_CONT " in ", regs->ip);
147 		pr_cont("\n");
148 	}
149 }
150 
151 static void
152 do_trap(int trapnr, int signr, char *str, struct pt_regs *regs,
153 	long error_code, int sicode, void __user *addr)
154 {
155 	struct task_struct *tsk = current;
156 
157 	if (!do_trap_no_signal(tsk, trapnr, str, regs, error_code))
158 		return;
159 
160 	show_signal(tsk, signr, "trap ", str, regs, error_code);
161 
162 	if (!sicode)
163 		force_sig(signr);
164 	else
165 		force_sig_fault(signr, sicode, addr);
166 }
167 NOKPROBE_SYMBOL(do_trap);
168 
169 static void do_error_trap(struct pt_regs *regs, long error_code, char *str,
170 	unsigned long trapnr, int signr, int sicode, void __user *addr)
171 {
172 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "entry code didn't wake RCU");
173 
174 	if (notify_die(DIE_TRAP, str, regs, error_code, trapnr, signr) !=
175 			NOTIFY_STOP) {
176 		cond_local_irq_enable(regs);
177 		do_trap(trapnr, signr, str, regs, error_code, sicode, addr);
178 		cond_local_irq_disable(regs);
179 	}
180 }
181 
182 /*
183  * Posix requires to provide the address of the faulting instruction for
184  * SIGILL (#UD) and SIGFPE (#DE) in the si_addr member of siginfo_t.
185  *
186  * This address is usually regs->ip, but when an uprobe moved the code out
187  * of line then regs->ip points to the XOL code which would confuse
188  * anything which analyzes the fault address vs. the unmodified binary. If
189  * a trap happened in XOL code then uprobe maps regs->ip back to the
190  * original instruction address.
191  */
192 static __always_inline void __user *error_get_trap_addr(struct pt_regs *regs)
193 {
194 	return (void __user *)uprobe_get_trap_addr(regs);
195 }
196 
197 DEFINE_IDTENTRY(exc_divide_error)
198 {
199 	do_error_trap(regs, 0, "divide error", X86_TRAP_DE, SIGFPE,
200 		      FPE_INTDIV, error_get_trap_addr(regs));
201 }
202 
203 DEFINE_IDTENTRY(exc_overflow)
204 {
205 	do_error_trap(regs, 0, "overflow", X86_TRAP_OF, SIGSEGV, 0, NULL);
206 }
207 
208 #ifdef CONFIG_X86_F00F_BUG
209 void handle_invalid_op(struct pt_regs *regs)
210 #else
211 static inline void handle_invalid_op(struct pt_regs *regs)
212 #endif
213 {
214 	do_error_trap(regs, 0, "invalid opcode", X86_TRAP_UD, SIGILL,
215 		      ILL_ILLOPN, error_get_trap_addr(regs));
216 }
217 
218 static noinstr bool handle_bug(struct pt_regs *regs)
219 {
220 	bool handled = false;
221 
222 	if (!is_valid_bugaddr(regs->ip))
223 		return handled;
224 
225 	/*
226 	 * All lies, just get the WARN/BUG out.
227 	 */
228 	instrumentation_begin();
229 	/*
230 	 * Since we're emulating a CALL with exceptions, restore the interrupt
231 	 * state to what it was at the exception site.
232 	 */
233 	if (regs->flags & X86_EFLAGS_IF)
234 		raw_local_irq_enable();
235 	if (report_bug(regs->ip, regs) == BUG_TRAP_TYPE_WARN) {
236 		regs->ip += LEN_UD2;
237 		handled = true;
238 	}
239 	if (regs->flags & X86_EFLAGS_IF)
240 		raw_local_irq_disable();
241 	instrumentation_end();
242 
243 	return handled;
244 }
245 
246 DEFINE_IDTENTRY_RAW(exc_invalid_op)
247 {
248 	irqentry_state_t state;
249 
250 	/*
251 	 * We use UD2 as a short encoding for 'CALL __WARN', as such
252 	 * handle it before exception entry to avoid recursive WARN
253 	 * in case exception entry is the one triggering WARNs.
254 	 */
255 	if (!user_mode(regs) && handle_bug(regs))
256 		return;
257 
258 	state = irqentry_enter(regs);
259 	instrumentation_begin();
260 	handle_invalid_op(regs);
261 	instrumentation_end();
262 	irqentry_exit(regs, state);
263 }
264 
265 DEFINE_IDTENTRY(exc_coproc_segment_overrun)
266 {
267 	do_error_trap(regs, 0, "coprocessor segment overrun",
268 		      X86_TRAP_OLD_MF, SIGFPE, 0, NULL);
269 }
270 
271 DEFINE_IDTENTRY_ERRORCODE(exc_invalid_tss)
272 {
273 	do_error_trap(regs, error_code, "invalid TSS", X86_TRAP_TS, SIGSEGV,
274 		      0, NULL);
275 }
276 
277 DEFINE_IDTENTRY_ERRORCODE(exc_segment_not_present)
278 {
279 	do_error_trap(regs, error_code, "segment not present", X86_TRAP_NP,
280 		      SIGBUS, 0, NULL);
281 }
282 
283 DEFINE_IDTENTRY_ERRORCODE(exc_stack_segment)
284 {
285 	do_error_trap(regs, error_code, "stack segment", X86_TRAP_SS, SIGBUS,
286 		      0, NULL);
287 }
288 
289 DEFINE_IDTENTRY_ERRORCODE(exc_alignment_check)
290 {
291 	char *str = "alignment check";
292 
293 	if (notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_AC, SIGBUS) == NOTIFY_STOP)
294 		return;
295 
296 	if (!user_mode(regs))
297 		die("Split lock detected\n", regs, error_code);
298 
299 	local_irq_enable();
300 
301 	if (handle_user_split_lock(regs, error_code))
302 		return;
303 
304 	do_trap(X86_TRAP_AC, SIGBUS, "alignment check", regs,
305 		error_code, BUS_ADRALN, NULL);
306 
307 	local_irq_disable();
308 }
309 
310 #ifdef CONFIG_VMAP_STACK
311 __visible void __noreturn handle_stack_overflow(const char *message,
312 						struct pt_regs *regs,
313 						unsigned long fault_address)
314 {
315 	printk(KERN_EMERG "BUG: stack guard page was hit at %p (stack is %p..%p)\n",
316 		 (void *)fault_address, current->stack,
317 		 (char *)current->stack + THREAD_SIZE - 1);
318 	die(message, regs, 0);
319 
320 	/* Be absolutely certain we don't return. */
321 	panic("%s", message);
322 }
323 #endif
324 
325 /*
326  * Runs on an IST stack for x86_64 and on a special task stack for x86_32.
327  *
328  * On x86_64, this is more or less a normal kernel entry.  Notwithstanding the
329  * SDM's warnings about double faults being unrecoverable, returning works as
330  * expected.  Presumably what the SDM actually means is that the CPU may get
331  * the register state wrong on entry, so returning could be a bad idea.
332  *
333  * Various CPU engineers have promised that double faults due to an IRET fault
334  * while the stack is read-only are, in fact, recoverable.
335  *
336  * On x86_32, this is entered through a task gate, and regs are synthesized
337  * from the TSS.  Returning is, in principle, okay, but changes to regs will
338  * be lost.  If, for some reason, we need to return to a context with modified
339  * regs, the shim code could be adjusted to synchronize the registers.
340  *
341  * The 32bit #DF shim provides CR2 already as an argument. On 64bit it needs
342  * to be read before doing anything else.
343  */
344 DEFINE_IDTENTRY_DF(exc_double_fault)
345 {
346 	static const char str[] = "double fault";
347 	struct task_struct *tsk = current;
348 
349 #ifdef CONFIG_VMAP_STACK
350 	unsigned long address = read_cr2();
351 #endif
352 
353 #ifdef CONFIG_X86_ESPFIX64
354 	extern unsigned char native_irq_return_iret[];
355 
356 	/*
357 	 * If IRET takes a non-IST fault on the espfix64 stack, then we
358 	 * end up promoting it to a doublefault.  In that case, take
359 	 * advantage of the fact that we're not using the normal (TSS.sp0)
360 	 * stack right now.  We can write a fake #GP(0) frame at TSS.sp0
361 	 * and then modify our own IRET frame so that, when we return,
362 	 * we land directly at the #GP(0) vector with the stack already
363 	 * set up according to its expectations.
364 	 *
365 	 * The net result is that our #GP handler will think that we
366 	 * entered from usermode with the bad user context.
367 	 *
368 	 * No need for nmi_enter() here because we don't use RCU.
369 	 */
370 	if (((long)regs->sp >> P4D_SHIFT) == ESPFIX_PGD_ENTRY &&
371 		regs->cs == __KERNEL_CS &&
372 		regs->ip == (unsigned long)native_irq_return_iret)
373 	{
374 		struct pt_regs *gpregs = (struct pt_regs *)this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
375 		unsigned long *p = (unsigned long *)regs->sp;
376 
377 		/*
378 		 * regs->sp points to the failing IRET frame on the
379 		 * ESPFIX64 stack.  Copy it to the entry stack.  This fills
380 		 * in gpregs->ss through gpregs->ip.
381 		 *
382 		 */
383 		gpregs->ip	= p[0];
384 		gpregs->cs	= p[1];
385 		gpregs->flags	= p[2];
386 		gpregs->sp	= p[3];
387 		gpregs->ss	= p[4];
388 		gpregs->orig_ax = 0;  /* Missing (lost) #GP error code */
389 
390 		/*
391 		 * Adjust our frame so that we return straight to the #GP
392 		 * vector with the expected RSP value.  This is safe because
393 		 * we won't enable interupts or schedule before we invoke
394 		 * general_protection, so nothing will clobber the stack
395 		 * frame we just set up.
396 		 *
397 		 * We will enter general_protection with kernel GSBASE,
398 		 * which is what the stub expects, given that the faulting
399 		 * RIP will be the IRET instruction.
400 		 */
401 		regs->ip = (unsigned long)asm_exc_general_protection;
402 		regs->sp = (unsigned long)&gpregs->orig_ax;
403 
404 		return;
405 	}
406 #endif
407 
408 	idtentry_enter_nmi(regs);
409 	instrumentation_begin();
410 	notify_die(DIE_TRAP, str, regs, error_code, X86_TRAP_DF, SIGSEGV);
411 
412 	tsk->thread.error_code = error_code;
413 	tsk->thread.trap_nr = X86_TRAP_DF;
414 
415 #ifdef CONFIG_VMAP_STACK
416 	/*
417 	 * If we overflow the stack into a guard page, the CPU will fail
418 	 * to deliver #PF and will send #DF instead.  Similarly, if we
419 	 * take any non-IST exception while too close to the bottom of
420 	 * the stack, the processor will get a page fault while
421 	 * delivering the exception and will generate a double fault.
422 	 *
423 	 * According to the SDM (footnote in 6.15 under "Interrupt 14 -
424 	 * Page-Fault Exception (#PF):
425 	 *
426 	 *   Processors update CR2 whenever a page fault is detected. If a
427 	 *   second page fault occurs while an earlier page fault is being
428 	 *   delivered, the faulting linear address of the second fault will
429 	 *   overwrite the contents of CR2 (replacing the previous
430 	 *   address). These updates to CR2 occur even if the page fault
431 	 *   results in a double fault or occurs during the delivery of a
432 	 *   double fault.
433 	 *
434 	 * The logic below has a small possibility of incorrectly diagnosing
435 	 * some errors as stack overflows.  For example, if the IDT or GDT
436 	 * gets corrupted such that #GP delivery fails due to a bad descriptor
437 	 * causing #GP and we hit this condition while CR2 coincidentally
438 	 * points to the stack guard page, we'll think we overflowed the
439 	 * stack.  Given that we're going to panic one way or another
440 	 * if this happens, this isn't necessarily worth fixing.
441 	 *
442 	 * If necessary, we could improve the test by only diagnosing
443 	 * a stack overflow if the saved RSP points within 47 bytes of
444 	 * the bottom of the stack: if RSP == tsk_stack + 48 and we
445 	 * take an exception, the stack is already aligned and there
446 	 * will be enough room SS, RSP, RFLAGS, CS, RIP, and a
447 	 * possible error code, so a stack overflow would *not* double
448 	 * fault.  With any less space left, exception delivery could
449 	 * fail, and, as a practical matter, we've overflowed the
450 	 * stack even if the actual trigger for the double fault was
451 	 * something else.
452 	 */
453 	if ((unsigned long)task_stack_page(tsk) - 1 - address < PAGE_SIZE) {
454 		handle_stack_overflow("kernel stack overflow (double-fault)",
455 				      regs, address);
456 	}
457 #endif
458 
459 	pr_emerg("PANIC: double fault, error_code: 0x%lx\n", error_code);
460 	die("double fault", regs, error_code);
461 	panic("Machine halted.");
462 	instrumentation_end();
463 }
464 
465 DEFINE_IDTENTRY(exc_bounds)
466 {
467 	if (notify_die(DIE_TRAP, "bounds", regs, 0,
468 			X86_TRAP_BR, SIGSEGV) == NOTIFY_STOP)
469 		return;
470 	cond_local_irq_enable(regs);
471 
472 	if (!user_mode(regs))
473 		die("bounds", regs, 0);
474 
475 	do_trap(X86_TRAP_BR, SIGSEGV, "bounds", regs, 0, 0, NULL);
476 
477 	cond_local_irq_disable(regs);
478 }
479 
480 enum kernel_gp_hint {
481 	GP_NO_HINT,
482 	GP_NON_CANONICAL,
483 	GP_CANONICAL
484 };
485 
486 /*
487  * When an uncaught #GP occurs, try to determine the memory address accessed by
488  * the instruction and return that address to the caller. Also, try to figure
489  * out whether any part of the access to that address was non-canonical.
490  */
491 static enum kernel_gp_hint get_kernel_gp_address(struct pt_regs *regs,
492 						 unsigned long *addr)
493 {
494 	u8 insn_buf[MAX_INSN_SIZE];
495 	struct insn insn;
496 
497 	if (copy_from_kernel_nofault(insn_buf, (void *)regs->ip,
498 			MAX_INSN_SIZE))
499 		return GP_NO_HINT;
500 
501 	kernel_insn_init(&insn, insn_buf, MAX_INSN_SIZE);
502 	insn_get_modrm(&insn);
503 	insn_get_sib(&insn);
504 
505 	*addr = (unsigned long)insn_get_addr_ref(&insn, regs);
506 	if (*addr == -1UL)
507 		return GP_NO_HINT;
508 
509 #ifdef CONFIG_X86_64
510 	/*
511 	 * Check that:
512 	 *  - the operand is not in the kernel half
513 	 *  - the last byte of the operand is not in the user canonical half
514 	 */
515 	if (*addr < ~__VIRTUAL_MASK &&
516 	    *addr + insn.opnd_bytes - 1 > __VIRTUAL_MASK)
517 		return GP_NON_CANONICAL;
518 #endif
519 
520 	return GP_CANONICAL;
521 }
522 
523 #define GPFSTR "general protection fault"
524 
525 DEFINE_IDTENTRY_ERRORCODE(exc_general_protection)
526 {
527 	char desc[sizeof(GPFSTR) + 50 + 2*sizeof(unsigned long) + 1] = GPFSTR;
528 	enum kernel_gp_hint hint = GP_NO_HINT;
529 	struct task_struct *tsk;
530 	unsigned long gp_addr;
531 	int ret;
532 
533 	cond_local_irq_enable(regs);
534 
535 	if (static_cpu_has(X86_FEATURE_UMIP)) {
536 		if (user_mode(regs) && fixup_umip_exception(regs))
537 			goto exit;
538 	}
539 
540 	if (v8086_mode(regs)) {
541 		local_irq_enable();
542 		handle_vm86_fault((struct kernel_vm86_regs *) regs, error_code);
543 		local_irq_disable();
544 		return;
545 	}
546 
547 	tsk = current;
548 
549 	if (user_mode(regs)) {
550 		tsk->thread.error_code = error_code;
551 		tsk->thread.trap_nr = X86_TRAP_GP;
552 
553 		show_signal(tsk, SIGSEGV, "", desc, regs, error_code);
554 		force_sig(SIGSEGV);
555 		goto exit;
556 	}
557 
558 	if (fixup_exception(regs, X86_TRAP_GP, error_code, 0))
559 		goto exit;
560 
561 	tsk->thread.error_code = error_code;
562 	tsk->thread.trap_nr = X86_TRAP_GP;
563 
564 	/*
565 	 * To be potentially processing a kprobe fault and to trust the result
566 	 * from kprobe_running(), we have to be non-preemptible.
567 	 */
568 	if (!preemptible() &&
569 	    kprobe_running() &&
570 	    kprobe_fault_handler(regs, X86_TRAP_GP))
571 		goto exit;
572 
573 	ret = notify_die(DIE_GPF, desc, regs, error_code, X86_TRAP_GP, SIGSEGV);
574 	if (ret == NOTIFY_STOP)
575 		goto exit;
576 
577 	if (error_code)
578 		snprintf(desc, sizeof(desc), "segment-related " GPFSTR);
579 	else
580 		hint = get_kernel_gp_address(regs, &gp_addr);
581 
582 	if (hint != GP_NO_HINT)
583 		snprintf(desc, sizeof(desc), GPFSTR ", %s 0x%lx",
584 			 (hint == GP_NON_CANONICAL) ? "probably for non-canonical address"
585 						    : "maybe for address",
586 			 gp_addr);
587 
588 	/*
589 	 * KASAN is interested only in the non-canonical case, clear it
590 	 * otherwise.
591 	 */
592 	if (hint != GP_NON_CANONICAL)
593 		gp_addr = 0;
594 
595 	die_addr(desc, regs, error_code, gp_addr);
596 
597 exit:
598 	cond_local_irq_disable(regs);
599 }
600 
601 static bool do_int3(struct pt_regs *regs)
602 {
603 	int res;
604 
605 #ifdef CONFIG_KGDB_LOW_LEVEL_TRAP
606 	if (kgdb_ll_trap(DIE_INT3, "int3", regs, 0, X86_TRAP_BP,
607 			 SIGTRAP) == NOTIFY_STOP)
608 		return true;
609 #endif /* CONFIG_KGDB_LOW_LEVEL_TRAP */
610 
611 #ifdef CONFIG_KPROBES
612 	if (kprobe_int3_handler(regs))
613 		return true;
614 #endif
615 	res = notify_die(DIE_INT3, "int3", regs, 0, X86_TRAP_BP, SIGTRAP);
616 
617 	return res == NOTIFY_STOP;
618 }
619 
620 static void do_int3_user(struct pt_regs *regs)
621 {
622 	if (do_int3(regs))
623 		return;
624 
625 	cond_local_irq_enable(regs);
626 	do_trap(X86_TRAP_BP, SIGTRAP, "int3", regs, 0, 0, NULL);
627 	cond_local_irq_disable(regs);
628 }
629 
630 DEFINE_IDTENTRY_RAW(exc_int3)
631 {
632 	/*
633 	 * poke_int3_handler() is completely self contained code; it does (and
634 	 * must) *NOT* call out to anything, lest it hits upon yet another
635 	 * INT3.
636 	 */
637 	if (poke_int3_handler(regs))
638 		return;
639 
640 	/*
641 	 * irqentry_enter_from_user_mode() uses static_branch_{,un}likely()
642 	 * and therefore can trigger INT3, hence poke_int3_handler() must
643 	 * be done before. If the entry came from kernel mode, then use
644 	 * nmi_enter() because the INT3 could have been hit in any context
645 	 * including NMI.
646 	 */
647 	if (user_mode(regs)) {
648 		irqentry_enter_from_user_mode(regs);
649 		instrumentation_begin();
650 		do_int3_user(regs);
651 		instrumentation_end();
652 		irqentry_exit_to_user_mode(regs);
653 	} else {
654 		bool irq_state = idtentry_enter_nmi(regs);
655 		instrumentation_begin();
656 		if (!do_int3(regs))
657 			die("int3", regs, 0);
658 		instrumentation_end();
659 		idtentry_exit_nmi(regs, irq_state);
660 	}
661 }
662 
663 #ifdef CONFIG_X86_64
664 /*
665  * Help handler running on a per-cpu (IST or entry trampoline) stack
666  * to switch to the normal thread stack if the interrupted code was in
667  * user mode. The actual stack switch is done in entry_64.S
668  */
669 asmlinkage __visible noinstr struct pt_regs *sync_regs(struct pt_regs *eregs)
670 {
671 	struct pt_regs *regs = (struct pt_regs *)this_cpu_read(cpu_current_top_of_stack) - 1;
672 	if (regs != eregs)
673 		*regs = *eregs;
674 	return regs;
675 }
676 
677 #ifdef CONFIG_AMD_MEM_ENCRYPT
678 asmlinkage __visible noinstr struct pt_regs *vc_switch_off_ist(struct pt_regs *regs)
679 {
680 	unsigned long sp, *stack;
681 	struct stack_info info;
682 	struct pt_regs *regs_ret;
683 
684 	/*
685 	 * In the SYSCALL entry path the RSP value comes from user-space - don't
686 	 * trust it and switch to the current kernel stack
687 	 */
688 	if (regs->ip >= (unsigned long)entry_SYSCALL_64 &&
689 	    regs->ip <  (unsigned long)entry_SYSCALL_64_safe_stack) {
690 		sp = this_cpu_read(cpu_current_top_of_stack);
691 		goto sync;
692 	}
693 
694 	/*
695 	 * From here on the RSP value is trusted. Now check whether entry
696 	 * happened from a safe stack. Not safe are the entry or unknown stacks,
697 	 * use the fall-back stack instead in this case.
698 	 */
699 	sp    = regs->sp;
700 	stack = (unsigned long *)sp;
701 
702 	if (!get_stack_info_noinstr(stack, current, &info) || info.type == STACK_TYPE_ENTRY ||
703 	    info.type >= STACK_TYPE_EXCEPTION_LAST)
704 		sp = __this_cpu_ist_top_va(VC2);
705 
706 sync:
707 	/*
708 	 * Found a safe stack - switch to it as if the entry didn't happen via
709 	 * IST stack. The code below only copies pt_regs, the real switch happens
710 	 * in assembly code.
711 	 */
712 	sp = ALIGN_DOWN(sp, 8) - sizeof(*regs_ret);
713 
714 	regs_ret = (struct pt_regs *)sp;
715 	*regs_ret = *regs;
716 
717 	return regs_ret;
718 }
719 #endif
720 
721 struct bad_iret_stack {
722 	void *error_entry_ret;
723 	struct pt_regs regs;
724 };
725 
726 asmlinkage __visible noinstr
727 struct bad_iret_stack *fixup_bad_iret(struct bad_iret_stack *s)
728 {
729 	/*
730 	 * This is called from entry_64.S early in handling a fault
731 	 * caused by a bad iret to user mode.  To handle the fault
732 	 * correctly, we want to move our stack frame to where it would
733 	 * be had we entered directly on the entry stack (rather than
734 	 * just below the IRET frame) and we want to pretend that the
735 	 * exception came from the IRET target.
736 	 */
737 	struct bad_iret_stack tmp, *new_stack =
738 		(struct bad_iret_stack *)__this_cpu_read(cpu_tss_rw.x86_tss.sp0) - 1;
739 
740 	/* Copy the IRET target to the temporary storage. */
741 	__memcpy(&tmp.regs.ip, (void *)s->regs.sp, 5*8);
742 
743 	/* Copy the remainder of the stack from the current stack. */
744 	__memcpy(&tmp, s, offsetof(struct bad_iret_stack, regs.ip));
745 
746 	/* Update the entry stack */
747 	__memcpy(new_stack, &tmp, sizeof(tmp));
748 
749 	BUG_ON(!user_mode(&new_stack->regs));
750 	return new_stack;
751 }
752 #endif
753 
754 static bool is_sysenter_singlestep(struct pt_regs *regs)
755 {
756 	/*
757 	 * We don't try for precision here.  If we're anywhere in the region of
758 	 * code that can be single-stepped in the SYSENTER entry path, then
759 	 * assume that this is a useless single-step trap due to SYSENTER
760 	 * being invoked with TF set.  (We don't know in advance exactly
761 	 * which instructions will be hit because BTF could plausibly
762 	 * be set.)
763 	 */
764 #ifdef CONFIG_X86_32
765 	return (regs->ip - (unsigned long)__begin_SYSENTER_singlestep_region) <
766 		(unsigned long)__end_SYSENTER_singlestep_region -
767 		(unsigned long)__begin_SYSENTER_singlestep_region;
768 #elif defined(CONFIG_IA32_EMULATION)
769 	return (regs->ip - (unsigned long)entry_SYSENTER_compat) <
770 		(unsigned long)__end_entry_SYSENTER_compat -
771 		(unsigned long)entry_SYSENTER_compat;
772 #else
773 	return false;
774 #endif
775 }
776 
777 static __always_inline unsigned long debug_read_clear_dr6(void)
778 {
779 	unsigned long dr6;
780 
781 	/*
782 	 * The Intel SDM says:
783 	 *
784 	 *   Certain debug exceptions may clear bits 0-3. The remaining
785 	 *   contents of the DR6 register are never cleared by the
786 	 *   processor. To avoid confusion in identifying debug
787 	 *   exceptions, debug handlers should clear the register before
788 	 *   returning to the interrupted task.
789 	 *
790 	 * Keep it simple: clear DR6 immediately.
791 	 */
792 	get_debugreg(dr6, 6);
793 	set_debugreg(DR6_RESERVED, 6);
794 	dr6 ^= DR6_RESERVED; /* Flip to positive polarity */
795 
796 	return dr6;
797 }
798 
799 /*
800  * Our handling of the processor debug registers is non-trivial.
801  * We do not clear them on entry and exit from the kernel. Therefore
802  * it is possible to get a watchpoint trap here from inside the kernel.
803  * However, the code in ./ptrace.c has ensured that the user can
804  * only set watchpoints on userspace addresses. Therefore the in-kernel
805  * watchpoint trap can only occur in code which is reading/writing
806  * from user space. Such code must not hold kernel locks (since it
807  * can equally take a page fault), therefore it is safe to call
808  * force_sig_info even though that claims and releases locks.
809  *
810  * Code in ./signal.c ensures that the debug control register
811  * is restored before we deliver any signal, and therefore that
812  * user code runs with the correct debug control register even though
813  * we clear it here.
814  *
815  * Being careful here means that we don't have to be as careful in a
816  * lot of more complicated places (task switching can be a bit lazy
817  * about restoring all the debug state, and ptrace doesn't have to
818  * find every occurrence of the TF bit that could be saved away even
819  * by user code)
820  *
821  * May run on IST stack.
822  */
823 
824 static bool notify_debug(struct pt_regs *regs, unsigned long *dr6)
825 {
826 	/*
827 	 * Notifiers will clear bits in @dr6 to indicate the event has been
828 	 * consumed - hw_breakpoint_handler(), single_stop_cont().
829 	 *
830 	 * Notifiers will set bits in @virtual_dr6 to indicate the desire
831 	 * for signals - ptrace_triggered(), kgdb_hw_overflow_handler().
832 	 */
833 	if (notify_die(DIE_DEBUG, "debug", regs, (long)dr6, 0, SIGTRAP) == NOTIFY_STOP)
834 		return true;
835 
836 	return false;
837 }
838 
839 static __always_inline void exc_debug_kernel(struct pt_regs *regs,
840 					     unsigned long dr6)
841 {
842 	/*
843 	 * Disable breakpoints during exception handling; recursive exceptions
844 	 * are exceedingly 'fun'.
845 	 *
846 	 * Since this function is NOKPROBE, and that also applies to
847 	 * HW_BREAKPOINT_X, we can't hit a breakpoint before this (XXX except a
848 	 * HW_BREAKPOINT_W on our stack)
849 	 *
850 	 * Entry text is excluded for HW_BP_X and cpu_entry_area, which
851 	 * includes the entry stack is excluded for everything.
852 	 */
853 	unsigned long dr7 = local_db_save();
854 	bool irq_state = idtentry_enter_nmi(regs);
855 	instrumentation_begin();
856 
857 	/*
858 	 * If something gets miswired and we end up here for a user mode
859 	 * #DB, we will malfunction.
860 	 */
861 	WARN_ON_ONCE(user_mode(regs));
862 
863 	if (test_thread_flag(TIF_BLOCKSTEP)) {
864 		/*
865 		 * The SDM says "The processor clears the BTF flag when it
866 		 * generates a debug exception." but PTRACE_BLOCKSTEP requested
867 		 * it for userspace, but we just took a kernel #DB, so re-set
868 		 * BTF.
869 		 */
870 		unsigned long debugctl;
871 
872 		rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
873 		debugctl |= DEBUGCTLMSR_BTF;
874 		wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl);
875 	}
876 
877 	/*
878 	 * Catch SYSENTER with TF set and clear DR_STEP. If this hit a
879 	 * watchpoint at the same time then that will still be handled.
880 	 */
881 	if ((dr6 & DR_STEP) && is_sysenter_singlestep(regs))
882 		dr6 &= ~DR_STEP;
883 
884 	if (kprobe_debug_handler(regs))
885 		goto out;
886 
887 	/*
888 	 * The kernel doesn't use INT1
889 	 */
890 	if (!dr6)
891 		goto out;
892 
893 	if (notify_debug(regs, &dr6))
894 		goto out;
895 
896 	/*
897 	 * The kernel doesn't use TF single-step outside of:
898 	 *
899 	 *  - Kprobes, consumed through kprobe_debug_handler()
900 	 *  - KGDB, consumed through notify_debug()
901 	 *
902 	 * So if we get here with DR_STEP set, something is wonky.
903 	 *
904 	 * A known way to trigger this is through QEMU's GDB stub,
905 	 * which leaks #DB into the guest and causes IST recursion.
906 	 */
907 	if (WARN_ON_ONCE(dr6 & DR_STEP))
908 		regs->flags &= ~X86_EFLAGS_TF;
909 out:
910 	instrumentation_end();
911 	idtentry_exit_nmi(regs, irq_state);
912 
913 	local_db_restore(dr7);
914 }
915 
916 static __always_inline void exc_debug_user(struct pt_regs *regs,
917 					   unsigned long dr6)
918 {
919 	bool icebp;
920 
921 	/*
922 	 * If something gets miswired and we end up here for a kernel mode
923 	 * #DB, we will malfunction.
924 	 */
925 	WARN_ON_ONCE(!user_mode(regs));
926 
927 	/*
928 	 * NB: We can't easily clear DR7 here because
929 	 * idtentry_exit_to_usermode() can invoke ptrace, schedule, access
930 	 * user memory, etc.  This means that a recursive #DB is possible.  If
931 	 * this happens, that #DB will hit exc_debug_kernel() and clear DR7.
932 	 * Since we're not on the IST stack right now, everything will be
933 	 * fine.
934 	 */
935 
936 	irqentry_enter_from_user_mode(regs);
937 	instrumentation_begin();
938 
939 	/*
940 	 * Start the virtual/ptrace DR6 value with just the DR_STEP mask
941 	 * of the real DR6. ptrace_triggered() will set the DR_TRAPn bits.
942 	 *
943 	 * Userspace expects DR_STEP to be visible in ptrace_get_debugreg(6)
944 	 * even if it is not the result of PTRACE_SINGLESTEP.
945 	 */
946 	current->thread.virtual_dr6 = (dr6 & DR_STEP);
947 
948 	/*
949 	 * The SDM says "The processor clears the BTF flag when it
950 	 * generates a debug exception."  Clear TIF_BLOCKSTEP to keep
951 	 * TIF_BLOCKSTEP in sync with the hardware BTF flag.
952 	 */
953 	clear_thread_flag(TIF_BLOCKSTEP);
954 
955 	/*
956 	 * If dr6 has no reason to give us about the origin of this trap,
957 	 * then it's very likely the result of an icebp/int01 trap.
958 	 * User wants a sigtrap for that.
959 	 */
960 	icebp = !dr6;
961 
962 	if (notify_debug(regs, &dr6))
963 		goto out;
964 
965 	/* It's safe to allow irq's after DR6 has been saved */
966 	local_irq_enable();
967 
968 	if (v8086_mode(regs)) {
969 		handle_vm86_trap((struct kernel_vm86_regs *)regs, 0, X86_TRAP_DB);
970 		goto out_irq;
971 	}
972 
973 	/* Add the virtual_dr6 bits for signals. */
974 	dr6 |= current->thread.virtual_dr6;
975 	if (dr6 & (DR_STEP | DR_TRAP_BITS) || icebp)
976 		send_sigtrap(regs, 0, get_si_code(dr6));
977 
978 out_irq:
979 	local_irq_disable();
980 out:
981 	instrumentation_end();
982 	irqentry_exit_to_user_mode(regs);
983 }
984 
985 #ifdef CONFIG_X86_64
986 /* IST stack entry */
987 DEFINE_IDTENTRY_DEBUG(exc_debug)
988 {
989 	exc_debug_kernel(regs, debug_read_clear_dr6());
990 }
991 
992 /* User entry, runs on regular task stack */
993 DEFINE_IDTENTRY_DEBUG_USER(exc_debug)
994 {
995 	exc_debug_user(regs, debug_read_clear_dr6());
996 }
997 #else
998 /* 32 bit does not have separate entry points. */
999 DEFINE_IDTENTRY_RAW(exc_debug)
1000 {
1001 	unsigned long dr6 = debug_read_clear_dr6();
1002 
1003 	if (user_mode(regs))
1004 		exc_debug_user(regs, dr6);
1005 	else
1006 		exc_debug_kernel(regs, dr6);
1007 }
1008 #endif
1009 
1010 /*
1011  * Note that we play around with the 'TS' bit in an attempt to get
1012  * the correct behaviour even in the presence of the asynchronous
1013  * IRQ13 behaviour
1014  */
1015 static void math_error(struct pt_regs *regs, int trapnr)
1016 {
1017 	struct task_struct *task = current;
1018 	struct fpu *fpu = &task->thread.fpu;
1019 	int si_code;
1020 	char *str = (trapnr == X86_TRAP_MF) ? "fpu exception" :
1021 						"simd exception";
1022 
1023 	cond_local_irq_enable(regs);
1024 
1025 	if (!user_mode(regs)) {
1026 		if (fixup_exception(regs, trapnr, 0, 0))
1027 			goto exit;
1028 
1029 		task->thread.error_code = 0;
1030 		task->thread.trap_nr = trapnr;
1031 
1032 		if (notify_die(DIE_TRAP, str, regs, 0, trapnr,
1033 			       SIGFPE) != NOTIFY_STOP)
1034 			die(str, regs, 0);
1035 		goto exit;
1036 	}
1037 
1038 	/*
1039 	 * Save the info for the exception handler and clear the error.
1040 	 */
1041 	fpu__save(fpu);
1042 
1043 	task->thread.trap_nr	= trapnr;
1044 	task->thread.error_code = 0;
1045 
1046 	si_code = fpu__exception_code(fpu, trapnr);
1047 	/* Retry when we get spurious exceptions: */
1048 	if (!si_code)
1049 		goto exit;
1050 
1051 	force_sig_fault(SIGFPE, si_code,
1052 			(void __user *)uprobe_get_trap_addr(regs));
1053 exit:
1054 	cond_local_irq_disable(regs);
1055 }
1056 
1057 DEFINE_IDTENTRY(exc_coprocessor_error)
1058 {
1059 	math_error(regs, X86_TRAP_MF);
1060 }
1061 
1062 DEFINE_IDTENTRY(exc_simd_coprocessor_error)
1063 {
1064 	if (IS_ENABLED(CONFIG_X86_INVD_BUG)) {
1065 		/* AMD 486 bug: INVD in CPL 0 raises #XF instead of #GP */
1066 		if (!static_cpu_has(X86_FEATURE_XMM)) {
1067 			__exc_general_protection(regs, 0);
1068 			return;
1069 		}
1070 	}
1071 	math_error(regs, X86_TRAP_XF);
1072 }
1073 
1074 DEFINE_IDTENTRY(exc_spurious_interrupt_bug)
1075 {
1076 	/*
1077 	 * This addresses a Pentium Pro Erratum:
1078 	 *
1079 	 * PROBLEM: If the APIC subsystem is configured in mixed mode with
1080 	 * Virtual Wire mode implemented through the local APIC, an
1081 	 * interrupt vector of 0Fh (Intel reserved encoding) may be
1082 	 * generated by the local APIC (Int 15).  This vector may be
1083 	 * generated upon receipt of a spurious interrupt (an interrupt
1084 	 * which is removed before the system receives the INTA sequence)
1085 	 * instead of the programmed 8259 spurious interrupt vector.
1086 	 *
1087 	 * IMPLICATION: The spurious interrupt vector programmed in the
1088 	 * 8259 is normally handled by an operating system's spurious
1089 	 * interrupt handler. However, a vector of 0Fh is unknown to some
1090 	 * operating systems, which would crash if this erratum occurred.
1091 	 *
1092 	 * In theory this could be limited to 32bit, but the handler is not
1093 	 * hurting and who knows which other CPUs suffer from this.
1094 	 */
1095 }
1096 
1097 DEFINE_IDTENTRY(exc_device_not_available)
1098 {
1099 	unsigned long cr0 = read_cr0();
1100 
1101 #ifdef CONFIG_MATH_EMULATION
1102 	if (!boot_cpu_has(X86_FEATURE_FPU) && (cr0 & X86_CR0_EM)) {
1103 		struct math_emu_info info = { };
1104 
1105 		cond_local_irq_enable(regs);
1106 
1107 		info.regs = regs;
1108 		math_emulate(&info);
1109 
1110 		cond_local_irq_disable(regs);
1111 		return;
1112 	}
1113 #endif
1114 
1115 	/* This should not happen. */
1116 	if (WARN(cr0 & X86_CR0_TS, "CR0.TS was set")) {
1117 		/* Try to fix it up and carry on. */
1118 		write_cr0(cr0 & ~X86_CR0_TS);
1119 	} else {
1120 		/*
1121 		 * Something terrible happened, and we're better off trying
1122 		 * to kill the task than getting stuck in a never-ending
1123 		 * loop of #NM faults.
1124 		 */
1125 		die("unexpected #NM exception", regs, 0);
1126 	}
1127 }
1128 
1129 #ifdef CONFIG_X86_32
1130 DEFINE_IDTENTRY_SW(iret_error)
1131 {
1132 	local_irq_enable();
1133 	if (notify_die(DIE_TRAP, "iret exception", regs, 0,
1134 			X86_TRAP_IRET, SIGILL) != NOTIFY_STOP) {
1135 		do_trap(X86_TRAP_IRET, SIGILL, "iret exception", regs, 0,
1136 			ILL_BADSTK, (void __user *)NULL);
1137 	}
1138 	local_irq_disable();
1139 }
1140 #endif
1141 
1142 void __init trap_init(void)
1143 {
1144 	/* Init cpu_entry_area before IST entries are set up */
1145 	setup_cpu_entry_areas();
1146 
1147 	/* Init GHCB memory pages when running as an SEV-ES guest */
1148 	sev_es_init_vc_handling();
1149 
1150 	idt_setup_traps();
1151 
1152 	/*
1153 	 * Should be a barrier for any external CPU state:
1154 	 */
1155 	cpu_init();
1156 
1157 	idt_setup_ist_traps();
1158 }
1159